20070402Application, Wind Integration Study.pdf

THE LAW OFFICE OF
PAINE , HAMBLEN LLP
R. Blair Strong
Partner
r. blair. strong(Q)painehamblen. com
717 WEST SPRAGUE AVENUE
SUITE 1200
SPOKANE, WASHINGTON 99201-3505
[phone]
FAX: [phone]
www.painehamblen.com
" ,, .',.." ..
:. Q.
",::
'leGl pJ'i ;( ..' 2 (,i \ ), J
March 30, 2007
Ms. Jean D. Jewell
Commission Secretary
Idaho Public Utilities Commission
472 West Washington
Boise, Idaho 83702
RE:Case No. AVU-o7-0~
In The Matter Of the Petition of A vista Corporation
Dear Ms. Jewell:
Enclosed for filing please find the original and seven copies of the Petition of A vista
Corporation. Attached to the Petition is the Final Report, A vista Corporation Wind Integration
Study. In its Petition, A vista requests various changes to its published avoided cost rates and
rules regarding purchases from qualifying facilities under the Public Utility Policies Regulatory
Act of 1978 and Commission Rules and Regulations.
If you should have any questions, please do not hesitate to contact me. Please conform
and return the additional copies in the enclosed self-addressed stamped envelope. Thank you for
your assistance.
Very truly yours
PAINE HAMBLEN LLP
??
73~ S)V~
R. Blair Strong
1:\Spodocs\1 1 150\04226\ltr\OO494550,DOC
Enclosure
Limited Liability PartnershiP
Offices in Spokane' Coeur d'Alene . Kennewick
c i
R. Blair Strong
Jerry K. Boyd (Idaho Bar #1341)
Paine Hamblen LLP
717 West Sprague, Suite 1200
Spokane, W A 99220-3505
Telephone: [phone]
Facsimile: 509-838-00007
?CiJ:t /.:
:, -
0" ') '-
",,--'"
'I
" """" ,, ::" !
,j j i"-- j i L
,', .." : ' " .,' '~ -'.
ii,'!;::; ~'ii
Attorneys for A vista Corporation
BEFORE THE IDAHO PUBLIC UTILITIES COMMISSION
IN THE MATTER OF THE PETITION OF
A VISTA CORPORATION FOR AN ORDER
REVISING AVISTA CORPORATION'
OBLIGATIONS TO ENTER INTO
CONTRACTS TO PURCHASE ENERGY
GENERATED BY WIND-POWERED SMALL)
POWER GENERATION FACILITIES
No. A VU-67-o-L...-
PETITION
Avista Corporation ("A vista ), pursuant to the Rule 31 , IDAPA 31.01.01.031 , hereby
petitions the Idaho Public Utilities Commission ("Commission ) to issue an Order:
Raising the cap on entitlement to published avoided cost rates for wind-powered small
power generation facilities that are qualifying facilities ("QFs ) under Sections 201 and 210 of
the Public Utility Regulatory Policies Act of 1978 ("PURPA") from the current level of 100 kW
to 10 average MWs per month ("10 aMW"), subject to the following conditions:
Reducing the published avoided cost rates applicable to purchases by
A vista of electric power from wind-powered QFs by 12%, as a percentage reduction to be
applied against scheduled avoided cost rates in those circumstance, except where the QF
developer agrees in the power purchase and sale contract with A vista to deliver QF
output to A vista on a firm hourly schedule, in which case the percentage reduction shall
be 6%;
PETITION -
Removing the requirement that the 90%/110% performance band
requirement not be applied to purchases from wind-powered QFs;
Authorizing A vista to purchase state-of-the-art wind forecasting services
to provide A vista with forecasted wind conditions in those geographic areas in which
wind generation resources are located, provided that QFs will reimburse A vista for their
share of the on-going cost of the wind forecasting service, in proportion to their
percentage share of the wind-generator capability being supplied to A vista from that area;
Requiring QFs to deliver a "mechanical availability guarantee" to A vista
to demonstrate monthly, except for scheduled maintenance and events of force majeure
or uncontrollable force, that the QF was physically capable and available to generate a
full output during 85% of the hours in a month;
Clarifying the rules governing the entitlement to published rates to prevent
all QFs, whether wind or non-wind, capable of delivering more than 10 aMW per month
from structuring or restructuring into smaller projects solely for the purpose to qualify for
the published avoided cost rates; and
Clarifying that the cap on entitlement to published avoided cost rates shall
be raised to 10 aMW only until A vista s total wind portfolio from all sources totals 400
MW.
BACKGROUND
Idaho Power Company filed a Petition with the Commission on June 17, 2005 , in Case
No. IPC-05-22 requesting that the Commission suspend its obligations under Sections 201 and
210 of PURP A to enter into new contracts to purchase energy generated by wind-powered QFs.
PETITION - 2
On June 28 2005, Avista filed a "Petition and Comments " in Case No. IPC-05-22 in which it
requested that the Commission temporarily suspend A vista s PURP A purchase obligations in the
same manner as requested by Idaho Power rather than temporarily suspending wind-powered
purchase obligations as requested by the Utilities. In Order No. 29839, issued on August 4
2005, the Commission reduced the published rate eligibility cap for QF wind projects from
10 aMW per month to 100 kW and required individual contract negotiations for QFs larger than
100 kW for A vista, Idaho Power and PacifiCorp.
Since the commencement of Case No. IPC-0522, Avista participated in follow-up wind
workshops in Idaho and initiated its own wind integration study. In particular, A vista retained a
leading industry consultant, EnerNex, to prepare a wind integration study, the final report of
which was released to the public on March 20, 2007. Additionally, A vista participated in, and
benefitted from Wind Integration Action Plan proceedings conducted under the auspices of the
Bonneville Power Administration and the Northwest Power & Conservation Council.
Recently, Idaho Power Company has filed Petitions respecting PURP A purchase
obligations in two different dockets, Case Nos. IPC-07-03 and IPC-07-04. The petition of
Avista, herein, in particular, items 1 , 2, 3, 4, 5, and 6, above, reflect recommendations made by
Idaho Power Company with respect to its service territory. A vista concurs with those
recommendations, and therefore recommends that similar policies be adopted with respect to
A vista s PURP A purchase obligations.
II.
INTEGRATION COSTS OF WIND-POWERE TO BE REFLECTED
IN DISCOUNT TO BE APPLIED TO PUBLISHED AVOIDED COST RATES
With respect to the costs of integrating wind generation into existing utility systems, the
Commission found in Order No. 29839, in Case No. IPC-05-, that the supply characteristics
PETITION - 3
of wind generation and related integration costs could provide a basis for adjustment of the
published avoided cost rates, an adjustment that may be different for each utility A vista
recommendation that published avoided cost rates applicable be discounted by 12% reflects
A vista s assessment, based on its wind integration study, of the cost to its system associated with
integrating up to 400 MW (nameplate capacity) of wind generation, both constructed and
purchased, into its system. The 12% discount reflects costs associated with integrating the QF
output into A vista s system, including intra-hour variability of deliveries from wind generation.
Where the QF developer agrees, as part of the power purchase and sale contract with
A vista to deliver QF output to A vista on a firm hourly schedule, A vista recommends that the
published avoided cost rates instead be discounted by 6%, which reflects a lower level of costs
incurred by A vista in integrating the wind power.
A vista submits herein as Exhibit A to this Petition, its Wind Integration Study. The study
indicates that for up to 400 MW (name plate capacity) the cost of integrating wind generated
energy into A vista s system is correlated to the purchase price for such energy. Additionally, the
Wind Integration Study supports A vista s recommendations, because of the intermittent nature of
wind, and its site specific location, that the costs of integrating up to 400 MW are approximately
12% more than the costs of integrating other non-wind resources that would be available to
A vista, including costs associated with intra-hour variability of wind. Where the wind-generated
power is delivered to A vista on a firm hourly schedule, the study shows that the integration costs
are approximately 6%.
Therefore, Avista recommends that a 12% discount be applied against the applicable
avoided cost rates set forth in A vista s rate schedules where the wind power is delivered to the
A vista system on a dynamic moment-to-moment basis. Where wind power is delivered on a firm
PETITION - 4
hourly schedule, Avista recommends a 6% discount. The 12% discount would be applied across
all otherwise applicable rates, and the developer would remain free to pick the duration of
contract, and levelized or unlevelized rates. Where the QF developer assumes the cost of
delivering a firm hourly product, only a 6% discount would be applied.
A vista recommends that the applicable discount be applied to the published avoided cost
rates to determine a purchase and sale price that will be established for the duration of the
contract. A vista recommends this approach in order to assure QFs that are of a size less than
10 aMW a predictable rate.
III.
ELIMINATION OF THE 90%/110% PERFORMANCE BAND
Idaho Power Company in Case No. IPC-07-03 recommended the elimination of the
90%/110% performance band, subject to several conditions. A vista recommends that the same
policies be applied to purchases of wind power by A vista from QFs. A vista believes that its
proposed discount captures, as best as can be determined presently, the cost of integrating wind
generation into A vista s system and, therefore to some degree, takes into account the inherent
difficulty of accurately forecasting the availability of wind. The establishment of the discount
will in large measure account for the variability of wind, and thereby diminish the need for a
performance band for wind.Furthermore, A vista believes there is benefit to a level of
consistency the in the structure of PURP A QF tariffs between utilities.
Additionally, A vista supports the concept of establishing a mechanical availability
guarantee grid. This guarantee would encourage wind developers to maintain the readiness of
their equipment throughout the full duration of the long term contract.
PETITION - 5
A vista also supports the concept that QFs should participate in funding wind forecasting
services, as a condition for not being bound by the performance band. Where such services are
purchased or constructed by A vista within a geographic area, A vista would propose to share such
expense on a pro rata basis with QFs that are selling their power to A vista under long-term
contracts, so that the QFs would pay a proportion of the wind forecasting expenses proportional
to their share of the wind-generator capability within the A vista wind portfolio from that
geographic region.
IV.
ADOPTION OF A RULE PREVENTING MULTIPLE PROJECTS
OWNED BY THE SAME PERSON FROM RECEIVING THE PUBLISHED
AVOIDED COST RATES, IF LOCATED IN THE SAME SITE
Idaho Power Company has recommended adoption of a rule nearly the same as that
adopted by the Oregon Public Utility Commission. A vista recommends that the approach
recommended by Idaho Power in Case No. IPC-07-04 be applied to Avista s purchases as well.
Wind projects are uniquely able to reconfigure themselves into various legal ownerships
solely for economic reasons, without disturbing or affecting in any way site or structural design.
In some circumstances, other generating technologies may have the similar capability. These
projects that are under common ownership should not be able to be reconfigure themselves
legally, for the sole purpose of qualifying for published avoided costs in Idaho.
Additionally, a uniform approach as between Idaho jurisdictional utilities is particularly
useful, in order to avoid unneeded incentives for favoring one utility over another, not because of
the fundamental economic differences reflected in the avoided costs and wind integration costs
but because of different QF rules that might apply to different utilities.
PETITION - 6
APPLICATION TO WIND GENERATION OF UP TO 400 MW
(NAME PLATE RATING) ON A VISTA'S SYSTEM
A vista recommends that the cap on entitlement to published avoided cost rates be raised
to 10 aMW only until Avista s total wind portfolio from all sources totals 400 MW, or until
A vista files for changes to its avoided cost schedules, or files a new Wind Integration Cost study
based on additional industry experience. The 400 MW represents the targeted amount of wind
power acquisition, based upon A vista s IRP. As part of its IRP process and other resource
assessment processes, A vista plans to continue studying the costs and effects of wind power
upon its system, and will make further filings, when it appears that material changes to its tariff
are necessary.
VI.
COMMUNICATIONS
Communications respecting this matter should be addressed to:
R. Blair Strong
Paine Hamblen LLP
717 W. Sprague Avenue, Suite 1200
Spokane, W A 99201-3505
Telephone: [phone]
Facsimile: [phone]
Clint Kalich
Manager of Resource Planning & Power
Supply Analyses
Avista Corporation
O. Box 3727
1411 East Mission Avenue, MSC- 7
Spokane, Washington 99220-3727
Telephone: [phone]
Facsimile: [phone]
VII.
MODIFIED PROCEDURE
Avista has sent via email to all the parties that participated in Case No. IPC-05-22 a
copy of the Petition. If no parties file comments on A vista s proposal, or indicate substantial
opposition to A vista s Petition in written comments, A vista requests that the Commission
PETITION - 7
consider this Petition in accordance with Rule 201 et seq.allowing for disposition by Modified
Procedure. IDAPA 31.01.01.201 et seq.A vista is receptive to further proceedings, if the
Commission believes, based on comments received, that further proceedings would be
advantageous.
WHEREFORE A vista respectfully petitions the Commission to issue an Order:
Raising the cap on entitlement to published avoided cost rates for wind-powered small
power generation facilities that are qualifying facilities ("QFs ) under Sections 201 and 210 of
the Public Utility Regulatory Policies Act of 1978 ("PURP A") from the current level of 100 kW
to 10 average MWs per month ("10 aMW"), subject to the following conditions:
Reducing the published avoided cost rates applicable to purchases by
Avista of electric power from wind-powered QFs by 12%, as a percentage reduction to be
applied against scheduled avoided cost rates in those circumstance, except where the QF
developer agrees in the power purchase and sale contract with A vista to deliver QF
output to A vista on a firm hourly schedule, in which case the percentage reduction shall
be 6%;
Removing the requirement that the 90%/110% performance band
requirement not be applied to purchases from wind-powered QFs;
Authorizing A vista to purchase state-of-the-art wind forecasting services
to provide A vista with forecasted wind conditions in those geographic areas in which
wind generation resources are located, provided that QFs will reimburse A vista for their
share of the on-going cost of the wind forecasting service, in proportion to their
percentage share of the wind-generator capability being supplied to A vista from that area;
PETITION - 8
Requiring QFs to deliver a "mechanical availability guarantee" to A vista
to demonstrate monthly, except for scheduled maintenance and events of force majeure
or uncontrollable force, that the QF was physically capable and available to generate a
full output during 85% of the hours in a month;
Clarifying the rules governing the entitlement to published rates to prevent
all QFs, whether wind or non-wind, capable of delivering more than 10 aMW per month
from structuring or restructuring into smaller projects solely for the purpose to qualify for
the published avoided cost rates; and
Clarifying that the cap on entitlement to published avoided cost rates shall
be raised to 10 aMW only until A vista s total wind portfolio from all sources totals
400 MW.
t-&..
RESPECTFULLY SUBMITTED this day of March, 2007.
PAINE HAMBLEN LLP
By:
/? ~-
S-t.
By:
I:\Spodocs\1 1 1 50\04 145\plead\00493996,DQC
PETITION - 9
f n e N e'x
., ,on ,
- ,- ''-"
l'r' ":;CJ
LCC;! i\i ~ -2 ;,; I .J'
,.:;
UT\L ri": ;r~"~ij:'
;\~'
A vu- c-o'l-O~
Final Report
Avista Corporation Wind Integration Study
Prepared for:
A vista Corporation
c/o Mr. Clint Kalich
Manager of Resource Planning
O. Box 3727
1411 East Mission Street
Spokane, Washington 99220-3727
Tel: [phone]
FAX: [phone]
clint .kalichCQJavistacorp.com
Prepared by:
EnerNex Corporation
1 70C Market Place Boulevard
Knoxville, Tennessee 37922
Tel: [phone] ext. 149
FAX: [phone]
bobzCQJenernex.com
wwvv.enernex.com
March, 2007
170C Market Place Boulevard. Knoxville, TN 37922 . Tel: [phone] . Fax: [phone]
www.enernex.com
Contents
EXECUTIVE SUMMARY
Higher Wind Penetration Equals Higher Integration Cost
Integration Costs Are Correlated with Market Prices
Shorter-Term Markets Can Reduce Cost of Variability
Rising Forecast Error Increases Integration Cost
Geographic Diversity has Direct Influence on Integration Costs
Operational Coordination between the Control Center and Wind Generators Can
Reduce Integration Costs
SECTION 1 OVERVIEW OF UTILITY SYSTEM OPERATIONS
Ancillary Services for Maintaining Power System Reliability and Security
Where do Ancillary Services "Come From
Measuring Power System Control Performance
Operational Planning
SECTION 2 WIND INTEGRATION STUDY METHODOLOGY
Study Methodology
Impacts of Wind Generation Within the Hour
SECTION 3 DEVELOPING THE WIND GENERATION MODEL
Wind Generation Model Characteristics
SECTION 4 OVERVIEW OF AVISTA SYSTEM OPERATION
Control Area Load
Control Area Resources
The Pacific Northwest Wholesale Marketplace
Overview of Resource Operations Theory
SECTION 5 MODELING AND ASSUMPTIONS
Overview of the Avista System Integration LP Model
The Controls Module
The Assumptions Module
The Clark Fork Logic and Mid-C Logic Modules
The Pre-Schedule Model and Real-Time Model Modules
Scenario Analysis and Objectives
Operation Of A Hydro-Based Control Area
SECTION 6 IMPACTS OF WIND GENERATION WITHIN THE HOUR
Modeling and Analysis for Wind Integration Assessment
Calculating Requirements for Managing Variability within the Hour
Impacts of Short-Term Forecast Error on Real-Time Operations
SECTION 7 RESULTS
f n e e'x
VII
xii
xii
xii
CORPORATION Page i
Base Case Results
Sensitivity Analysis
Impact of Market Structure
Value of Wind Curtailment
The Value ofImproved Wind Generation Forecasting
Integration Cost Sensitivity to Market Conditions
Impact of Reduced Forecast Error
Benefits of Geographical Diversity
SECTION 8 SUMMARY
Higher Wind Penetration Equals Higher Integration Cost
Integration Costs Are Correlated with Market Prices
Shorter-Term Markets Can Reduce Cost of Variability
Rising Forecast Error Increases Integration Cost
Geographic Diversity has Direct Influence on Integration Costs
Operational Coordination between the Control Center and Wind Generators Can
Reduce Integration Costs
SECTION 9 REFERENCES
APPENDIX A WIND GENERATION CONTRIBUTION TO PLANNING MARGIN
Background
Data and Analytical Method
Results
Application to Avista Resource Planning
APPENDIX B ADDITIONAL CHARTS AND TABLES
APPENDIX C NEXT STEPS
f n e e'XCORPORATION Page ji
Figures
Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
Figure 7:
Figure 8:
Figure 9:
Figure 10:
Figure 11:
Figure 12:
Figure 13:
Figure 14:
Figure 15:
Figure 16:
Figure 17:
Figure 18:
Figure 19:
Figure 20:
Figure 21:
Figure 22:
Figure 23:
Figure 24:
Figure 25:
Figure 26:
Figure 27:
Figure 28:
Figure 29:
t n e N e'XCORPORATION
Schematic of Avista LP program - Pre-schedule module
High wind period (3 days) showing load and wind generation by scenario.
Wind generation "penetration duration" curves for four wind scenarios.
NERC reliability regions and control areas
NERC Interconnected Operations Services
NERC CPS2 equations
Turbine power curve used in wind speed data conversion
Low wind period (3 days) showing load and wind generation by scenario.
High wind period (3 days) showing load and wind generation by scenario.
Production distribution for 100 MW scenario.
Production distribution for 200 MW scenario.
viii
Production distribution for 400 MW scenario.
Production distribution for 600 MW scenario.
Production duration curves for four wind scenarios.
Wind generation "penetration duration" curves for four wind scenarios. Avista control area resources
Schematic of Avista LP program - Pre-schedule module
Avista LP Model control interface
Mid-C Logic Module Maximizing Hydro Facility Value
Hourly average and ten-minute load
Hourly average and ten minute values, with over-the-hour ramp period
Hourly load, ten-minute load, and load following "requirement"
Ten-minute average load shown with up/down intra-hourly load following capability37
Variability of wind generation over one hour from LP Model data (by scenario)
Approximation of empirical wind generation variability with quadratic expressions
Ten-minute load net wind generation and intra-hourly load following capability
Actual and forecast hourly average values. Short-term forecast is made 1.5 hours
prior to the start of the subject hour.
Additional intra-hour flexibility requirements due to schedule error bias.
Page iii
Figure 30: Standard deviation of persistence forecast error over a two hour horizon for the fourwind generation scenarios.
Figure 31: Integration cost as function of short-term wind generation forecast error for base
scenarios.
Figure 33:
Figure 34:
Effects of geographic dispersion of wind generation facilities on integration cost. Basecase scenarios are indicated by the star symbols.
Average Summer and Winter ELCC Contributions
Winter ELCC Contribution Distributions 1989-2004 Figure 32:
t n eCORPORATION Page
Tables
Table 1:
Table 2:
Table 3:
Table 4:
Table 5:
Table 6:
Table 7:
Table 8:
Table 9:
Table 10:
Table 11:
Table 12:
Table 13:
Table 14:
Table 15:
Table 16:
Table 17:
Table 18:
Table 19:
Table 20:
Table 21:
Table 22:
Table 23:
Table 24:
Table 25:
Table 26:
Table 27:
f n e e'XCORPORATION
Annual Capacity Factor by Scenario (from LP Model data)
Total Reserves for Variability and Schedule Deviations
Integration Costs for Base Scenarios (33% capacity factor)
2006 CPS2 Bounds for some Western Interconnection Control Areas
Measurement Locations and Record Durations from OSU Wind Speed Database
utilized for study.
Composition of 100 MW Scenarios
Composition of 200 MW Scenarios
Composition of 400 MW Scenarios
Composition of 600 MW Scenarios
Annual Capacity Factor by Scenario (from LP Model data)
Avista 2006 Monthly Peak Control Area Demand
Illustration of plant capacity utilization
Avista LP Model Resource Assumptions (two tables)
Avista LP Model Transmission Assumptions
Total Operating Reserve Assumptions for Wind Scenarios
Average Hourly Flexibility Requirements for Managing Control Area Variability
Total Reserves for Variability and Schedule Deviations
Integration Costs for Base Scenarios
Incremental Reserve Assumptions for Base Scenarios
Components of Wind Integration Cost - Dollars
Components of Integration Cost - Percent
Hydroelectric Generation Portion of Integration Costs
Impact of Hydro Conditions on Integration Cost
Effect on Integration Cost of Short-Term Liquid Markets
Impact of Limited Wind Curtailment on Integration Cost
Integration Costs with Perfect Day-Ahead Forecast (no pre-schedule penalty)
Market Price Impacts on Integration Cost
viii
Page v
~ n e e'XCORPORATION Page vi
EXECUTIVE Su M MARY
The variability and uncertainty of wind energy production require that power system
operators take measures to manage its delivery. These measures may increase the cost
incurred to balance the system and maintain reliability. Over the past nine years , a
number of investigations have been conducted by electric utility organizations across
the country and around the world to better characterize and quantify the impacts of
wind generation on the operation of the grid.
This report documents an analysis conducted by Avista Corporation to quantify the
incremental costs to operations associated with integrating wind generation into its
control area. Four levels of wind generation were studied: 100 MW, 200 MW, 400 MW
and 600 MW. These generation levels are equivalent to 5% to 30% of control area peak
load. EnerNex Corporation of Knoxville , Tennessee was retained by Avista to guide in
the construction and application of a methodology which has been used in many of the
previous U.S. studies.
Avista s present work builds on analyses completed in 2001/2002. A proprietary Avista
System Integration LP Dispatch Model ("Avista LP Model", or "LP Model"), driven by a
linear programming engine , optimizes operations with and without wind generation in
the utility s system (Figure 1). This hourly LP Model tracks various capabilities (e., up
and down load following, regulation , energy, storage) of Avista s system to meet system
loads at least cost. It contains four modules. The first two optimize hydro generation
oil a daily basis at the Mid-Columbia and Clark Fork projects, tracking constraints
such as maximum and minimum storage and generation levels , and minimum flow. A
third creates the hourly pre-schedule, taking daily hydro quantities and allocating them
across the highest value hours possible given the remaining system constraints. The
pre-schedule LP Model contains day-ahead forecasts of load and wind generation.
Purchases and sales made to balance system requirements are carried forward to the
real-time module. The real-time module re-optimizes utility resources given the new
forecasts for wind and load. It performs tasks similar to the pre-schedule module.
The analysis of wind generation is based on simulating Avista s short-tenn scheduling
and dispatch operations over an extended chronological period. The primary inputs to
this simulation process are chronological profiles of system load, wind generation, and
market prices for energy purchases and sales. Load and market price data are
extracted from archives, but acquiring the wind generation data is much more
challenging. Recent studies show that a high-fidelity, long-term , chronological
representation of wind generation is the most critical study element. For large wind
generation development scenarios , it is very important that the effects of spatial and
geographic diversity be neither under- nor over-estimated.
f n e e'XCORPORATION Page vii
Figure 1:Schematic of Avista LP program - Pre-schedule module
The long-tenn wind speed data base compiled by Oregon State University s (OSU)
Energy Resources Research Laboratory (ERRL) was used as the basis for the
chronological wind generation model. Specifically, data from the five historical
Bonneville Power Administration (BPA) sites, along with observation3 from the operating
wind plant at Vansycle, were selected as the reference data points. Using methods and
algorithms developed in earlier studies, EnerNex utilized the wind speed data to
generate high-resolution wind energy production profiles. Annual chronological records
of wind generation at a-minute and hourly resolutions (Figure 2) were used in analysis
of wind generation impacts on Avista real-time operations, and were a critical input to
the annual dispatch simulations from which integration costs were derived.
Annual capacity factors for the four base wind generation scenarios are documented in
Table 1. These capacity factors were adjusted to a consistent 33% in the final
calculations of wind integration costs presented later in this study. The portion of
Avista load being served by wind generation over each hour of the study year is shown
for each scenario in Figure 3.
Table 1:Annual Capacity Factor by Scenario (from LP Model data)
' "'..'" '
~c~l1ari()"
.'...~.
Ul1adjpstedCapacify Fador
100MW 34.
200 MW 33.
400 MW 30.
600 MW 30.
t n e e'XCORPOR,.,TION Page viii
Figure 2:
Figure 3:
lCC'C'
--'
2::-:'C'
10':-:'
,-.--....--.. ""''" -',---'.-- -
, l~'C'C'i i i ,. I
nl'ln , r
, l
?\j """
--",,,u,,u '- &CO
.._,'._
J..
\.....
~\l\n.._rr,, rl\w1 \i\J~j"l,.j~~~u Un
' """ '
"i/II ',f:
,:
i '
' " . \ ~, ': "
II II"
, ~, \
i ,
~~"
'-"f"-J~;I,~A t'
: "
iI,
/'"
0 !i '
~(
: I
: :~~,
,"I'L
3249 3255 3~",1 3267 3273 3279 32",5 3291 3~97 ",2,C'3 33'09 3315 12,
..2
Hour
High wind period (3 days) showing load and wind generation by scenario.
"'0
..2
;::...
J::
100 MW
200 MW
----- 400
600 MW
"'0
0...
0, ,
..--'-, -,- '
1000 2000 3000 4000 5000 6000 7000 8000
# of Hours
Wind generation "penetration duration" curves for four wind scenarios.
Annual dispatch simulations were conducted using hourly load and wind generation
data, as is the norm for these types of studies. Consequently, it was first necessary to
assess how wind generation would impact Avista operations inside the hour. Here
generation is continuously adjusted to balance load and obligations for out-of-area
transactions. Additionally, geT).eration capacity must be held in reserve to cover sudden
losses of other generation or transmission facilities. The variability of wind generation
f n e e'xCORPORATION Page ix
will increase the level of variation already seen and managed by power system
operators, so the amount of capacity available to operators to balance the control area
is increased. The objective of the intra-hour analysis was to quantify this increase in
reserve capacity so that it could be represented in the hourly dispatch simulations.
Various mathematical and statistical analyses were used to quantify these impacts.
From high-resolution load and wind generation data, the amount of additional
generation capacity needed to manage the system in real time were extracted by
applying algorithms used in previous studies and some new approaches developed for
specific operating practices prevalent in the Pacific Northwest. Results of the analysis
(documented in detail in a later section) are shown in Table 2.
Table 2:Total Reserves for Variability and Schedule Deviations
!~rr~i~~~~~~r!
i~rii~t~~~~i\~~~~~~1;~~i;I~~~~r~~~~i:~:~i~~1~
Load only 20MW 35.0 MW
100MW 22.1 MW 38.3 MW
200 MW 24.1 MW 49.5 MW
400 MW 27.9 MW 68..7 MW
600 MW 31.0MW 103..7MW
Reserve requirements for load alone and the various levels of wind generation are
brought forward as inputs to the dispatch simulations. Because the time step is
hourly, the generation movements to balance load inside the hour are not actually
simulated directly. Instead, they become constraints on the scheduling and economic
dispatch algorithms , and increase costs over the course of the simulation since those
capacity amounts cannot be used to serve load or cover out-of-area sales.
While there is no formal definition for "integration cost", in the context of this study and
the others performed over the past several years it is the reduction in value of wind
energy due to its variability and uncertainty. So, to quantify integration cost, metrics
from the simulated scheduling and dispatch of actual wind generation (from the LP
Model) are compared to those of a resource that delivers an equivalent amount of energy
but has no variability and can be forecast perfectly.
Integration costs for the four base wind generation scenarios are shown in Table 3. It
should be noted that integration costs are functions of a large number of factors, and
changes to anyone of those assumptions could change the results. The results
presented here, then, must also be viewed in the context of the assumptions made, the
composition of Avista resource portfolio, and the rules and policies by which utilities
operating in the Pacific Northwest currently abide.
That said, the results are consistent with findings of a number of studies conducted
over the past four years. The 600 MW scenario represents a level of wind generation
penetration (30%) at or above the highest level which has been studied in detail by
earlier studies.
t n e e'XCORPORATION Page x
Table 3:Integration Costs for Base Scenarios (33% capacity factor)
.. Winc1
"::.
'JYinct Sy,sterrt i ForE:!cast :,
:'
Cost' ..
' "
Cost
. ,:..
tocatroOi ..Capacify,PenetratioIJ;JrroJt (~/MWh).
..:.
C%Mktl:
Columbia Basin 100MW 15%$2.75
50/50 Mix of CB & MT 200 MW 10%10%$6.12.7%
Diversified Mix 400 MW 20%$6.12.
Diversified Mix 600 MW 30%$8.16.
Beyond the results for the base scenarios, the sensitivity of computed integration costs
to a number of assumptions was evaluated through additional annual dispatch
simulations. A number of observations and conclusions were drawn from these cases;
they are reported qualitatively in the following paragraphs, and in the Results section of
this report.
HIGHER WIND PENETRATION EQUALS HIGHER INTEGRATION COST
The Avista study confirms what other studies before it have theorized or shown through
analysis. Higher wind penetration levels, all other things being equal, increase wind
integration costs. To provide a full understanding of wind integration costs, this study
ran the LP Model through varying levels of wind penetration, from five percent up to
approximately thirty percent. This wide range covers where many systems are today,
and pushes the envelope well beyond the 20% level cited as the point below which wind
can be accommodated with only modest cost impacts.
INTEGRATION COSTS ARE CORRELATED WITH MARKET PRICES
Capacity opportunity costs are a significant component of wind integration. As prices
rise , all things equal, one might expect integration costs to rise as well. Wind resource
value , therefore , does not rise equally with the market price, as integration costs
consume some of the additional value. Avista used the LP Model to look at two price
sensitivities - market prices equal to half of forecasted levels, and twice forecasted
levels - and found that market prices and wind integration costs -are correlated.
SHORTER- TERM MARKETS CAN REDUCE COST OF VARIABILITY
In this study, the increased short-term uncertainty due to wind generation forecast
errors increased the amount of reserve capacity required to operate the system. Much
of this is driven by rules that govern short-term exchanges of energy in the Pacific
Northwest. Because the "window" for hourly trading closes well in advance of the hour
probable errors in wind generation forecasts become significant.
While improvements in wind generation forecasting can assist, reduction of the lead
time for energy transactions would also have an influence. In regions with well-
functioning short-term energy markets (some cleared at intervals as short as 5
minutes), variability in demand due to both wind generation and load variability is
spread out over a much larger footprint. When the aggregation effects on variability
over this larger geographical area are considered, the net effects on system operation
can be substantially reduced.
f n r N e"xCORPORATION Page
RISING FORECAST ERROR INCREASES INTEGRATION COST
Forecast error affects the overall level of reserve capacity necessary to integrate wind
resources. As forecast error rises , so do integration costs. Many participants to the
wind integration debate disagree on how accurate wind forecasts, and hence forecast
error, are. This study strives to identify an appropriate level of reserves to account for
forecast error; the debate will continue. To this end, Avista ran its LP Model under
various levels of forecast error, from zero percent, or perfect foresight, to thirty percent.
GEOGRAPHIC DIVERSITY HAS DIRECT INFLUENCE ON INTEGRATION COSTS
Additional generation capacity must be reserved to manage increased control area
variability and uncertainty. This capacity is a major component of integration cost.
Wind plants concentrated in a small region will exhibit a much higher degree of
correlation in their output than plants separated by larger geographic distances.
OPERATIONAL COORDINATION BETWEEN THE CONTROL CENTER AND WIND
GENERATORS CAN REDUCE INTEGRATION COSTS
Impacts of wind generation variability and uncertainty on the control area are not
evenly distributed over all hours of the year. There can be times where the incremental
cost for managing wind generation rise dramatically. In these times, the most economic
solution may be to "feather" wind energy via production curtailments.
f n e e'xCORPORATION Page xii
Section 1
OVERVIEW OF UTILITY SYSTEM OPERATIONS
Interconnected power systems are large and extremely complex machines, consisting of
tens of thousands of individual elements. The mechanisms responsible for their control
must continually adjust the supply of electric energy to meet the combined and ever-
changing electric demand of the system s users. There are a host of constraints and
objectives that govern how this is done. For example, the system strives to operate with
very high reliability and provide electric energy at the lowest possible cost. The
operational limitations of individual network elements-generators, transmission lines
substations - must be honored at all times. The capabilities of each of these elements
must be utilized in a fashion to provide the required high levels of performance and
reliability at the lowest overall cost.
Operating the power system involves more than adjusting the combined output of
supply resources to meet the load. Maintaining reliability and acceptable performance
requires operators to:
Keep the voltage at each node (a point where two or more system elements -
lines, transformers, loads, generators, etc. - connect) of the system within
prescribed limits;
Regulate the system frequency (the steady electrical speed at which all
generators in the system are rotating) to keep all generating units in
synchronism;
Maintain the system in a state where it is able to withstand and recover from
unplanned failures or losses of major elements
Frequency and voltage are the fundamental performance indices for the system. High
interconnected power system reliability is a consequence of maintaining the system in a
secure state - a state where the loss of any element will not lead to cascading outages of
other equipment - at all times.
The electric power system in the United States (contiguous 48 states) is comprised of
three interconnected networks: the Eastern Interconnection (most of the states East of
the Rocky Mountains), the Western Interconnection (Rocky Mountain States west to the
Pacific Ocean), and ERCOT (most of Texas). Within the Eastern and Western
interconnections , dozens of individual "control" areas coordinate their activities to
maintain reliability and conduct transactions of electric energy with each other.
number of these individual control areas are members of Regional Reliability
Organizations (RROs), which oversee and coordinate activities across a number of
control areas to maintain the security of the interconnected power system.
A control area consists of generators, loads , and defined and monitored transmission
ties to neighboring areas. The Federal Energy Regulatory Commission definition of a
control area is:
t n e e)(CORPORATION Page
An electric power system or combination of electric power systems to which
common automatic control scheme is applied in order to: (1) match, at all times, the
power output of the generators within the electric power system(s) and capacity
and energy purchased from entities outside the electric power system(s), with the
load in the electric power system(s); (2) maintain, within the limits of Good Utility
Practice, scheduled interchange with other Control Areas; (3) maintain the
frequency of the electric power system(s) within reasonable limits in accordance
with Good Utility Practice; and (4) provide sufficient generating capacity
maintain operating reserves in accordance with Good Utility Practice.
Each control area must assist the larger interconnection with maintaining frequency at
60 Hz, and balance load, generation, out-of-area purchases and sales on a continuous
basis. In addition, a prescribed amount of backup or reserve capacity (generation that
is unused but available within a certain amount of time) must be maintained at all
times as protection against unplanned failure or outage of equipment.
To accomplish the objectives of minimizing costs and ensuring system performance and
reliability over the short term (hours to weeks), the activities that go on in each control
area consist of:
Developing plans and schedules for meeting the forecast load over the coming
days, weeks , and possibly months , considering all technical constraints
contractual obligations , and financial objectives;
Monitoring the operation of the control area in real time and making
adjustments when the actual conditions - load levels, status of generating units
etc. - deviate from the forecast.
ANCILLARY SERVICES FOR MAINTAINING POWER SYSTEM RELIABILITY AND
SECURITY
The activities and functions necessary for maintaining system performance and
reliability and minimizing costs are generally classified as "ancillary services." While
there is no universal agreement on the number or specific definition of these services
the following items encompass the range of technical aspects that must be considered
for reliable operation of the system:
Voltage regulation and VAR dispatch - deploying devices capable of generating
reactive power to manage voltages at all points in the network;
Regulation - the process of maintaining system frequency by adjusting certain
generating units in response to fast fluctuations in the total system load;
Load following - moving generation up (in the morning) or down (late in the day)
in response to daily load patterns;
Frequency-responding spinning reserve - maintaining an adequate supply of
generating capacity (usually on-line , synchronized to the grid) that is' able to
quickly respond to the loss of a major transmission network element or
generating unit;
Supplemental Reserve - managing an additional back-up supply of generating
capacity that can be brought on line relatively quickly to serve load in case of
the unplanned loss of significant operating generation or a major transmission
element.
f n e e'xCORPORATION Page
The nature of control area operations in real-time or in planning for the hours and days
ahead is such that increased knowledge of what will happen correlates strongly to
better strategies for managing the system. Much of this process is already based on
predictions of uncertain quantities. Hour- by- hour forecasts of load for the next day or
several days are critical inputs to the process of deploying electric generating units and
scheduling their operation. While it is recognized that load forecasts for future periods
can never be 100% accurate, they nonetheless are the foundation for all of the
procedures and processes for operating the power system. Increasingly sophisticated
load forecasting techniques and decades of experience in applying this information have
done much to lessen the effects of the inherent uncertainty
WHERE DO ANCILLARY SERVICES "COME FROM
Meeting the operational objectives for the power system is accomplished through
coordinated control of individual generators, the transmission network, and associated
auxiliary equipment such as shunt capacitor banks.
How individual plants are deployed and scheduled is primarily a function of economics.
Historically, vertically-integrated electric utilities would schedule their generating assets
to minimize total production costs for the forecast load while observing any constraintson the operation of the generating units in their fleet. In bulk power markets
competitive bidding either partially or wholly supplants the top-down optimization
performed by vertically-integrated utilities. In either case , the economics of unit power
production have the primary influence on how a plant is scheduled.
In addition, the entity responsible for the operation of the control area - an individual
utility or a regional transmission organization - must manage some generating units to
regulate frequency and control power exchanges in real time, to make up discrepancies
between actual and forecast loads, and provide adequate reserves to cover an
unexpected loss of supply.
The efficiency of thermal generating units typically varies with loading, so for each unit
there is a point at which the energy cost is minimized. For large fossil-fired and nuclear
generating units, the cost of generation generally declines with increasing loading up to
rated output. As a result, economics dictate that these units be "base loaded" for as
many hours as possible when in operation.Other factors , such as thermodynamic
system time constants or mechanical and thermal stresses may also result in certain
units being loaded at fairly constant levels.
Against these operating constraints for certain units, other generating resources are
deployed and scheduled to not only produce electric energy but also to provide the
flexibility necessary to regulate system frequency, follow the aggregate system load as it
trends up in the morning and down late in the day, and provide reserve capacity in the
case of a generating unit or tie line failure. Some of these functions are under the
auspices of a central, hierarchical control system generally referred to as automatic
generation control or AGC. Others involve human intervention by the control areaoperators. In either case , the generating units participating in the system control
activities must:
!The term "base loaded" is generally used to describe the operation of large generating units with high
capital and operating costs but low fuel costs that are loaded to near maximum capability for most of
the hours they are in service. In traditional electric utility system planning, the "base load" is
sometimes defined as the minimum hourly system demand over the course of a year.
f n e e'XCORPORATION Page 3
Be responsive to commands issued by the control area EMS (energy
management system), otherwise known as "being on AGC". Participating in AGC
generally requires a specific infrastructure for communications with control
center SCADA (furstem ~ontrol gnd .Qata Acquisition) system.
Operate with appropriate "head room" to increase or decrease generation
without violating minimum loading limits if commanded by the system operator
or energy management system.
Be able to change their output (move up or down, or "ramp ) quickly enough to
provide the required system regulation
The EMS is the technical core of modern control areas. It consists of hardware
software, communications , and telemetry to monitor the real-time performance of the
control area and make adjustments to generating unit and other network components
to achieve operating performance objectives. A number of these adjustments happen
very quickly without the intervention of human operators. Others are made in response
to decisions by individuals charged with monitoring system performance.
MEASURING POWER SYSTEM CONTROL PERFORMANCE
Control of the interconnected electric power systems in the U.S. is affected by the
coordinated actions of over 100 individual control centers. Figure 4 shows the NERC
(North American Electric Reliability Council) and the control areas within each region.
Within each control area, the supply of electric energy is continuously adjusted to
balance the requirements of loads and to maintain scheduled sales or purchases of
energy from other control areas.
The primary objective of the individual control centers is to operate the power system to
ensure security and reliability. Specific obligations contributing to this objective
include:
Meeting instantaneous demand, Interchange Schedule, Operating Reserve , and
reactive resource requirements.
Providing frequency bias obligations.
Balancing Net Actual Interchange and Net Scheduled Interchange
Using tie-line bias control (unless doing so would be adverse to system or
interconnection reliability).
Complying with Control Performance and Disturbance Control Standards
Repaying its Inadvertent Interchange balance.
It is interesting to note that there is no hierarchical control scheme - each control area
follows the same rules , but does not receive signals from a "master" controller charged
with the operation of the entire interconnection.
As defined by NERC, the activities and functions traditionally known as ancillary
services are called "Interconnected Operations Services (IOS)". Figure 5 illustrates the
three sub-categories of IOS. The services shown here are all provided through the
scheduling and control of generating resources.
~ n eCORPORATION Page 4
" "'--
MAAC
ECAR
, "' '
SERC
-- - - - - -..
DynamICally Cc,ntrc.Ded Generation ERCOT
Figure 4: NERC reliability regions and control areas
Interconnected
Operations
Services
Resource and
Demand
Balance
Bulk
Transmission
Reliability
Emergency
Preparedness
Regulation Reactive Power
Supply from
Generation
SourcesLoad Following
Contingency
Reserve
~ning
L ~~~PPlemental
Frequency
Response
System Control
Figure 5: NERC Interconnected Operations Services
f n e N e'x
As "I JaJ1uary 1 , 2005
Reliability
Objec6ves
IDS Building
Blocks
IDS
Deployment
CORPORATION Page 5
The fundamental quantity upon which generation control is based is known as Area
Control Error, or ACE.
Where
NIA =
NIs
p =
FA =
Fs =
IME =
ACE = (NIA - NIs) - lOB (FA - Fs) -IME
the sum of the actual interchange with other control areas
the total scheduled interchange with other control areas
the control area frequency bias , reflecting the fact that load will change
with frequency
the actual frequency of the interconnection
the scheduled frequency of the interconnection; this is usually 60 Hz
although there are times when the scheduled frequency is slightly above
or below the nominal value to affect what is know as "time error
correction
metering error, which will be neglected for the purposes of this
discussion
ACE is computed automatically by the control area EMS every few seconds. The
adequacy of generation adjustments by the control area operators and the EMS are
gauged by two metrics that use ACE as an input. The first metric, Control Performance
Standard 1 (CPSI), uses ACE values averaged over a 1 minute period. It is a measure
of how the control area is helping to support and manage the frequency of the entire
interconnection. If the interconnection frequency is low, it signifies that there is more
demand than generation (the "machine" is slowing down ). If a particular control area
has a negative ACE, it is contributing to this frequency depression. Conversely, if ACE
were positive during that period, over-generation in the control area is helping to restore
the interconnect frequency.
The CPSI "score" for control areas is based on performance over a rolling I2-month
period. This score must be greater than 100% (an artifact of the equations used to
compute the compliance factor). Maintaining adequate capacity on automatic
generation control is a major factor in complying with CPS 1. On the other hand, very
high CPSI scores can be an indication of over-control, which costs money and is not
required.
The second metric is Control Perfonnance Standard 2 (CPS2). It utilizes the average of
ten consecutive I-minute ACE values. Over each ten minute period, the ten-minute
average ACE for a control area must be within specific bounds, known as 110. These
bounds are unique for each control area and are based generally on system size. 2006
CPS2 bounds for selected control areas in the Western Interconnection are shown in
Table 4.
The CPS2 metric is tabulated monthly. To comply with CPS2 requirements , 90% or
more of the ten- minute average ACE values must be within the designated 110 bounds
for the control area. Minimum perfonnance allows 14.4 violations per day. Most
control areas keep their CPS2 scores in the mid 90% range.
f n eCORPORATION Page 6
Table 4:2006 CPS2 Bounds for some Western Interconnection Control Areas
2006 CPS2 Bounds
Est Pellk Freq. BillS Bills/Load Bias/Total l10 Variable
Demllnd (MW)(MW/.1Hz)('!oj Bills('!o)(MW)Bills?
WEcc-t/WPP
Alb~rta electric Syslem Cperater 980 114 1..14 58,
,ista Corp.132 21.3 1.00 1.G4 25,
;!onnev;Ue Power Administraton 039 110- no"67.VariabJe
Sr.tisn Columbia Transmission Corporabcn "A~2 114 - 2EO"1..03 59.Variab!e
Idaho ?ower Company 3A45 1..45 38,52
~lorthWeslern Energy 549 I 9 1.23 23..74
?aeiScorp-E..!137 1..01 46,
?acificorp-Wesl 959 ..a8 1.37 44.
Porland Geno!'31 "lecT.e Company 000 1.25 38,
PUD No, 1 of CheJan County 514 1.22 27,
PUD No, 1 of Douglas County 291 2.41 14.
?UD No, 2 of Granl County 550 1.22 27.
Pugel Sound ""ergy Q34 1..01 38,
SeattJe Departmen! of Ligh1ing 1..750 -40 2.27 1.96 34.
Sie"a Pa.,;fie Po.....er Company 955 19.1.00 24,
Tacoma PO'Ner 973 1..85 23,
Western Area Po'Ner Admin;S!,a"on - Lipper 115 1.74
Great Plains Wesl
WEcc-NWPP Totals:1197 851 1.31 41.62
Source: ftp:/ /www.nerc.com/pub/sys/all updl/oc/opman/CPS2Bounds 2006.pdf
AVGID-minule(ACE)
::;
CPS2 = I L - Violations"~",h * 100(Total Periodsm,"" - Unavailable Periods~"'h
Figure 6: NERC CPS2 equations
Control area compliance with NERC performance standards is defined as a combination
of CPS1 and CPS2 scores:
In compliance: CPSI ;,. 100%and CPS2 ;,. 90%
Out of compliance: CPS1 oe;: 100%CPS2 oe;: 90%
Maintaining compliance with the NERC control performance standards requires
maneuverable generating capacity to be available and controlled or dispatched to
compensate for fluctuations in control area demand. How much additional capacity is
necessary to maintain compliance as the amount of wind generation in the control area
grows is an obvious question. Wind g~neration exhibits variations over the range of
time frames relevant to control perfonnance, at least theoretically, increases the
requirement for IOS. Since IOS do not directly generate revenue, dedicating additional
capacity for these functions comes at a price to the control area operator.
OPERATIONAL PLANNING
Electric utilities use sophisticated strategies and tools for deploying their generating
resources to serve load reliably and at the lowest cost. Demand forecasts over the next
day to several days are the starting point for optimization processes that determine
which resources should be committed to operation, and how they should be scheduled
~ n e N e'xCORPORATION Page 7
to serve forecast load. The control and reliability needs of the system, along with
limitations of the generating units themselves, constrain this optimization problem.
Wind generation variability and uncertainty complicates this problem in various ways:
Short-term variations in wind generation (minutes to tens of minutes) can
necessitate the reservation of additional generating capacity to compensate for
excesses or deficiencies in the supply as the system load varies. In general, this
reserved capacity cannot be used to serve load.
Wind generation varies with meteorological patterns. These patterns usually do
not align with the daily load patterns. Wind plant production may be low during
the late afternoon, when daily load is at its highest, or may be high during the
overnight hours when the load is near daily minimums and the value of energy
is the lowest.
Errors in wind generation forecasts can increase the overall uncertainty for unit
commitment and scheduling. Since the operations plan is optimized using
forecast data, actual load and wind generation that significantly depart from the
forecast will cause the plan to be less then optimal , implying that the cost to
serve the load will be higher.
De'leloping plans and schedules involves evaluating a very large number of possibilities
for the deployment of generating resources. A major objective here is to utilize the
supply resources so that all obligations are met and the total cost to serve the projected
load is minimized. With a large number of individual generating units with many
different operational characteristics and constraints, and other supply options such as
energy purchases from other control areas, software tools must be employed to develop
optimal plans and schedules. These tools assist operators in making decisions to
commit" generating units for operation , since many units cannot realistically be
stopped or started at will. They are also used to develop schedules for the next day or
days that will result in minimum costs if the load forecasts are accurate.
t n e e'xCORPORATION Page 8
Section 2
WIND INTEGRATION STUDY METHODOLOGY
Avista s present work builds on analyses completed in 2001/2002. A proprietary
dispatch LP Model ("Avista LP Model", or LP Model"), driven by a linear programming
engine, optimizes operations with and without wind generation in the utility's system.
This hourly LP Model tracks various capabilities (e., up and down load following,
regulation, energy, storage) of Avista s system to meet system loads at least cost. It
contains three modules. The first optimizes hydro generation on a daily basis at the
Mid-Columbia and Clark Fork projects, tracking constraints such as maximum and
minimum storage and generation levels, and minimum flow. A second creates the
hourly pre-schedule , taking daily hydro quantities and allocating them across the
highest value hours possible given the remaining system constraints. The pre-schedule
LP Model contains day-ahead forecasts of load and wind generation. Purchases and
sales made to balance system requirements are carried forward to the real-time module.
The real-time module re-optimizes utility resources given the new forecasts for wind and
load. It perfonns tasks similar to the pre-schedule module.
The key cost driver is incremental reserves necessary to integrate wind into a utility
system. Reserve obligations were calculated by using historical utility data from 2002
through 2004. Specifically, regulation (up to 1 minute), load following (1 minute to one
hour), spinning and non-spinning operating reserves , and forecast error are input in the
Avista LP Model as constraints on system optimization. In the with-wind cases, reserve
quantities are necessary to cover incremental regulation, load following and forecast
error. No additional spinning or non-spinning reserves are assumed; these products
are tied to system load rather than generation plant operations. This assumption is not
the rule today in the Northwest, but this approach will be implemented in the nearfuture.
Incremental regulation and load following reserves are calculated first by identifying
levels necessary to meet load variability alone. A second step perfonns the same
analysis but nets wind generation against load when performing the calculations. In
each of these analyses, reserve levels ensure a 95% probability of meeting each period'
reserve obligation. This level exceeds current CPS1 and CPS2 requirements, whereby
control area operators must adhere to a 90% level. Differences between the two
analyses (with and without wind) identify the incremental reserve obligations included
in the with-wind scenarios.
Incremental regulation reserves necessary to integrate wind were found to be constant
across all hours, rising with the level of wind in Avista s control area. Load following
obligations varied both with the level of wind in Avista s control area and as hourly wind
generation levels changed. Each of these reserve products are met with spinning-
capable resources.
Forecast error, a product covered by reserving system capacity, was a significant focus
of the Avista study. Two-hour-ahead wind forecasts were compared to actual wind
generation levels. Forecast error was calculated at the 95% confidence interval and
t n eCORPORATION Page 9
carried across all hours in the up and down directions. Forecast error is met with
spinning-capable resources.
Avista considered various levels of wind from 100 MW to 600 MW, or between 5% and
30% of control area peak demand. Wind resources were evaluated in the Columbia
Basin, in Eastern Montana, as a 50%/50% mix of Columbia Basin and Eastern
Montana wind, and as a multi-state "diversified" mix with many smaller sites combined.
The diversified sites had significantly lower reserve obligations and costs when
compared to single basin resources. Wind generation data for the 2002 through 2004
calendar years was obtained from the 10-minute Bonneville Power Administration Long-
Term Wind Database of wind speed data. Data limitations required the analysis to
focus on the period from August 2002 through July 2003.
STUDY METHODOLOGY
There is no formal or rigorous definition of "integration cost." It is a term used to
describe the financial impact of wind generation variability and uncertainty on the
control area charged with accepting it. The term applies to the operational time frame
comprising the real-time management of conventional generating units and the short-
term planning for demand over the coming day or days.
A chronological operations simulation methodology is the standard analytical approach
for wind integration studies. This framework utilizes synchronized hourly load and
wind generation patterns. It mimics the scheduling and real-time operation activities
for the company or area of interest.
The methodology for the analysis is designed to quantify the costs of wind generation
variability and uncertainty in the operational time frame. These costs are assessed by
comparing operation costs for managing wind to one where the same amount of energy
is delivered by an ideal resource - one that imposes no incremental burden on
scheduling or real-time operations.
The ideal resource over the year is represented by a 12X24 shape. For each month a
unique 24-hour shape is used for every day. The 24-hour shape was calculated as the
average generation delivered during each separate clock hour of each month. The
second run incorporates actual-i.e., hourly variable-wind output and the required
additional operating reserves necessary to maintain a consistent level of system control
performance (CPSI & CPS2).
Wind integration cost is calculated as the difference between system values from each
run The difference between the two runs is divided by the total wind energy produced
during the year. This process is completed for each wind penetration level, wind
source , and water year, and evaluated scenario.
IMPACTS OF WIND GENERATION WITHIN THE HOUR
The main objective of this study is to determine how the Avista control area would be
impacted by the additions of wind generation. An analysis combining Avista load and
simulated wind generation data determines the requirements for regulation and load
following necessary to maintain system reliability. The findings from the load and wind
analysis become inputs to later analytical processes. An LP Model developed by Avista
takes these data and simulates the operational changes necessary to provide thesecapabilities.
f n e e'xCORPORATION Page 10
The approach for analyzing intra-hour wind generation impacts is based on
straightforward mathematical and statistical analyses using ten-minute averages of
system load and wind generation. Incremental reserve requirements are determined by
comparing various metrics of the load by itself(and the present capacity and
capabilities allocated to perform these services) to the combination of load and wind
generation.
While the load exhibits a larger trend pattern of steadily increasing over the morning
interval and falling in the evening, there is still significant variability in the load by
itself. Deviations in wind generation are less patterned than load variability.
f n eCORPORATION Page 1/
Section 3
DEVELOPING THE WIND GENERATION MODEL
The analysis of wind generation is based on simulating Avista s short-term scheduling
and dispatch operations over an extended chronological period. The primary inputs to
this simulation process are chronological profiles of system load, wind generation, and
market prices for energy purchases and sales. Load and market price data are
extracted from archives, but acquiring the wind generation data is much more
problematic. Recent studies show that a high-fidelity, long-term, chronological
representation of wind generation is the most critical study element. For large wind
generation development scenarios, it is very important that the effects of spatial and
geographic diversity be neither under- nor over-estimated.
The long-term wind speed data base compiled by Oregon State University's (OSU)
Energy Resources Research Laboratory (ERRL) was used as the basis for the
chronological wind generation model Specifically, data from the five historical
Bonneville Power Administration (BPA) sites, along with observations from the operating
wind plant at Vansyc1e, were selected as the reference data points. The historical
period of data utilized from each of these sites is shown in Table
Table 5:Measurement Locations and Record Durations from OSU Wind Speed Database utilized
for study.
" .. .' ,. ,
sitEi~ame
...
firstD~te'/
..::
.i,
...
ii
.."
.iLastD
~te
Browning Depot, MT 1/29/20026:40:00 AM 8/13/2003 2:50:00 AM
Cape Blanco, OR 1/28/20024:40:00 PM 9/10/2003 1 :50:00 AM
Kennewick, WA 1/29/20026:30:00 AM 12/31/2004 11 :50:00 PM
Goodnoe Hills, WA 1/29/20026:00:00 AM 12/31/200411:50:00 PM
Sevenmile Hill, OR 1/29/20026:00:00 AM 12/31/2004 11 :50:00 PM
Vansycle, OR 8/2/2002 3:20:00 PM 12/31/200411:50:00 PM
Wind speed data at each location was transformed into wind energy production at 10-
minute intervals using a power curve from a commercially available 2.75 MW wind
turbine (NEG 2750). The power curve for this turbine is shown in Figure 7. The
selection of power curve is not critical, since the objective is to create a long-term
pattern of wind generation. Production over an extended period is a secondary issue.
f n e e'xCORPORATION Page
3000
2500
.,:::;::-
2000
:;....
..::L
....
1500
...,..
1000
500
0 Q
Wind Speed (m/s)
Figure 7:Turbine power curve used in wind speed data conversion
There are a number of factors influencing aggregate production from a collection of
wind turbines in a given area. Measurement data collected by the National Renewable
Energy Laboratory and others over the past decade provides empirical data on the effect
of aggregation and spatial diversity on production variability. As the number of
turbines in the collection grows, the aggregate production begins to smooth , first at
small time scales (seconds to minutes), then progressively over long spans of time.
the wind production is spread over distinct geographical locations, production can be
significantly smoothed over spans of even multiple hours.
With the data used for this study, it is difficult to account for spatial diversity; there is a
single observation point of wind speed at each location. Some smoothing is introduced
in the algorithm for translation from wind speed to production. This method has been
employed, and validated, in previous studies
The net effect of the data limitations is that individual wind generation profiles in this
study exhibit more variability than actual wind plants constructed at those locations.
To minimize the potential for increased generation variability and its potential for
biasing study results, four wind generation scenarios ranging from 100 MW to 600 MW
were constructed using wind speed data from the five sites in the OSU database and
observations from Vansyde (Table 6 through Table 9).
f n e e'XCORPORATION Page
Table 6:Composition of 100 MW Scenarios
Browning Depot, MT
Goodnoe Hills
Cape Blanco
Vansycle
Kennewick
Seven mile
100
Table 7:Composition of 200 MW Scenarios
.....'..".'....'
~it~;~~~n i'.'
(:"..'
;~i
MW~;\ih .i~0~j~~~~a
:jirr
..'. ..,
~~1~.,' i"
....
Browning Depot, MT
Goodnoe Hills
Cape Blanco
Vansycle
Kennewick
Sevenmile
100
100
200
Table 8:Composition of 400 MW Scenarios
..'..,'' "
:i"
.,.:
.O,,mvEirse:,
,.,
:, GolumbiaBasirr'
, ,,.., ,
Montana-
('rn'F)'F? (MW~c.;.i , (MVVy ' (IIJIWl:',i" "'
Browning Depot, MT
Goodnoe Hills
Cape Blanco
Vansycle
Kennewick
Seven mile
125
150
200
200
400
Table 9:Composition of 600 MW Scenarios
, "
.Divers iaB sin'Montan
~rnE!:./'d
'."
(MW)i,.i:
..,.,": '
(MIJV~..'
..,
" i,
.,,"'
,'." (MW)\,
Browning Depot, MT
Goodnoe Hills
Cape Blanco
Vansycle
Kennewick
Seven mile
195
225
300
300
600
Given the issues related to the exaggerated variability of the LP Model mentioned above
care was taken to reduce the effect on the integration impacts and costs to be
calculated later in the project by appropriately selecting the scenarios for each level of
wind generation to be studied. For example, a 600 MW scenario from a single location
(e.g. Montana) was not considered to be realistic due to the additional variability.
~ n e e'xCORPORATION Page
WIND GENERATION MODEL CHARACTERISTICS
The wind generation model consists of ten-minute data for extended chronological
periods and is therefore difficult to examine directly. The following charts and graphs
are intended to convey certain of these characteristics. Figure 8 and Figure 9 illustrate
two three-day periods of Avista load and wind generation from each scenario. The
aforementioned issue of higher wind generation variability is apparent during the high
wind generation periods , especially for the 600 MW scenario. Compared with the 400
MW scenario which uses the same locations but includes a "build out" of each, some
smoothing between these two scenarios would be expected, but is not evident from the
plots.
Figure 8:
:.=
oJ)
oJ)
1C-:0
30J
Wind Gen. -100 MW
- "
Nind Gen. - 200 I\-\"N
Wind Gen. - 400
Wind Gen. - 6C-O MW . 16-:0
Load
6CO-"-.. i-'--
---
.v:.o. ..m.
---! ,., ,----_
J..
---
T---' ..
,..-------'
20".0
2cok---
' "-"""'"'-
' ..m_--.n
';"~
.v:.o
("i'I\~ ,.1, "(f\r'"
0 ":l~:~,.~JF'~J~~~==="
,y,~,~,\,~,,:..-=-'
(~h,Jrr~?12;; 1261 1267 1273 127'1 123; 1291 1297 1303 1",~9 131; '13,1327
Hour
Low wind period (3 days) showing load and wind generation by scenario.
f n e e)(
2'X'J
....!
CORPORATION Page 15
l,:O:Q
~:Q '
---~'----
I -
!.-- -----
l-...---
+----+-,- ...
1 ~C'J
= -
A A
! -
, L
- ,
\ L
- ~~
li':I()tl \1
! '
V\ /\1
z:'J
~~\
I~,
j ~,
:t~\ -:~fA
~~~
/v~t
\/'
~t1 J~:t~~ m ~l"
...,.. --..--
i--, --...
-. 4I;QI ,I.'
! :
, i " I . i ,
: ""
i :'
, ,, '-'
I ~r !~\i/ i
" '
l-A~--;.~~~., i \I'~ I I
i -""--'1 ;",~I
~::.....
I I
! --...: !' ,- - ... : .
7 '"
3249 3255 3261 3267 3273 3279 3285 3291 :;'297 :;'3.:)3 33oJ9 :;.315 3:;'21
::::0:':'
VVind Gen. - 100 ,lAW
VVind Gen. - 2CO ,'vIW
:;:::
....J
0:(
Hour
Figure 9:High wind period (3 days) showing load and wind generation by scenario.
Table 12 shows the computed annual capacity factor for each wind generation scenario.
Figure 10 through 9 document the hourly production distributions for the four
scenarios. Note that as additional sites are added to the mix - i.e. all scenarios except
100 MW - the aggregate production falls short of the nameplate capacity. Production
duration curves for the year of data used in this analysis are shown in Figure 14.
Table 10:Annual Capacity Factor by Scenario (from LP Model data)
, ,..
ScenariO'Unadjusted Capacity Fador
100MW 34.
200 MW 33.
400 MW 30.
600 MW 30.
t n e e'XCORPORATION Page 16
Figure 10:
Figure 11:
~ n e e'XCORPORATION
::J
:r:
Hourly Production (MW)
Production distribution for 100 MW scenario.
::J
:r:
100 120 140 160
Hourly Production (MW)
Production distribution for 200 MW scenario.
180
100
200
Page
Figure 1
Figure 13:
t n e e'xCORPORATION
(j)
120 160 200 240
Hourly Production (MW)
Production distribution for 400 MW scenario.
(j)
120 180 240 300 360
Hourly Production (MW)
Production distribution for 600 MW scenario.
280
420
320
480
360
540
400
600
Page
100MW
200 MW
---- 400 M
600 MW
, '-
0....
" -- '," -, ..,'--..-
n ~n
~ -~-----"""-------:;,..-.::
..c
1500 3000 4500
~ '~---------'-----~,..
n_._
6000 7500 9000
# of Hours
Figure 14: Production duration curves for four wind scenarios.
In Figure 15, the hourly fraction of wind generation relative to load is calculated and
sorted to show the number of hours over the year where the wind to load fraction is
above the amount on the horizontal axis.
--.--- 100
200 MW
------ 400 M
600MW I
5000 8000
0....
----------. "
'm ~ ~_um
6000 7000
- - -
1000 2000 3000 4000
Figure 15: Wind generation "penetration duration" curves for four wind scenarios.
# of Hours
( n e e'XCORPORATION Page
Secti 0 n
OVERVIEW OF AVISTA SYSTEM OPERATION
Avista is a vertically-integrated natural gas and electricity company providing service to
350 000 electricity customers in the states ofIdaho and Washington. Avista also
provides control area services to a number of smaller external customers, including
large industrial facilities and small municipally-owned electric systems.
CONTROL AREA LOAD
In 2006 Avista recorded a control area peak demand of approximately 2,100 MW. The
minimum control area demand was approximately 890 MW. Avista is a winter-peaking
system, with peak winter loads exceeding peak summer loads by approximately ten to
fifteen percent. Table 11 provides 2006 monthly control area peak demand, as well as
average and minimum load levels.
Table 11:Avista 2006 Monthly Peak Control Area Demand
, ,."
Month
. ,
Min:
' ,
Max
..'
Average
CONTROL AREA RESOURCES
1109
1126
1071
952
963
932
979
975
892
951
978
1103
1820
2082
1799
1580
1761
1904
2021
1850
1711
1795
2110
1950
1477
1534
1417
1268
1277
1296
1421
1356
1237
1297
1455
1587
Avista relies on approximately 2 800 megawatts of owned or contracted resources to
serve the needs of its control area. In addition to serving control area load, the
Company also is obligated to provide approximately 200 megawatts to third-party
control areas. The following figure provides a summary of these resources.
f n e e'XCORPORATION Page 20
Figure 16: Avista control area resources
Avista uses a variety of resources to meet its overall load obligations; the majority of
capacity reserves are met by hydroelectric plants. Hourly schedules "block load" all
non-hydro resources in most hours , leaving hydro units with the responsibility to cover
intra-hour ramps regulation and load following in most hours. Other system operations
are possible; however, the Company has found that firing gas generation, for example
results in higher reserve carrying costs , Integrating wind into Avista s system does not
affect this relationship. The LP Model confirmed the economics of this mode of
operation.
THE PACIFIC NORTHWEST WHOLESALE MARKETPLACE
Utilities in the Northwest benefit from a robust wholesale marketplace, both on the day-
ahead pre-schedule time frame and the next-hour real-time period. Many reserve
products are available through short-term bilateral contracts. For many years Avista
has operated its resources around the market availability of third-party resources in
this marketplace. Where market prices are lower in a given hour than the cost of utility-
owned or controlled resources, purchases are made to serve control area obligations.
Where resources in excess of control area needs can be operated for a cost less than the
wholesale market price, this excess is marketed to the benefit of Avista s retail
customers.
Modeling the wholesale marketplace i~ essential for studying wind integration in the
Northwest. Unlike some large systems in North America, Avista s costs are not
necessarily driven by its system marginal cost. Instead, the wholesale marketplace
determines this cost by reflecting the marginal cost of the entire integrated system.
With respect to wind, its value in the operations timeframe is equal or nearly equal to
the short-tenn wholesale market price for power.
f n e e'XCORPORATION Page 2
OVERVIEW OF RESOURCE OPERATIONS THEORY
Utilities have a fiduciary responsibility to optimize resource operations in a least-cost
and reliable manner, Given a set of generation assets and load obligations , the two
should be matched in the most efficient manner. Generation assets enable a utility to
create various power products necessary to meet customer requirements. Amongthese
are energy, regulation, spinning and non-spinning reserves. Each resource has a
unique mix of abilities to provide these products. For example, a flexible hydroelectric
generator on automatic generation control may be able to provide all of these services
where a nuclear plant can only provide energy.
In addition to physical limitations a specific generation unit might have, it cannot create
power products beyond its capacity rating. A 100 MW plant can create 100 MW of
energy or follow 100 MW of increasing load, but not both. Table 12 provides examples
of the power products three hypothetical 100 MW generators might generate over an
hour in a market where energy, regulation, load following, and non-spinning reserves
are demanded. Notice that in each case the non-energy products produced never
exceed rated capacity. Capacity in the "down" direction cannot exceed energy
generation levels and "" direction capacity cannot exceed the difference between the
nameplate rating of the plant and the energy generation level.
Table 12:Illustration of plant capacity utilization
.."
NUdeOfi;;/
..,.'.'
..coor
.;'
' Hydro'
.."
Energy 100 100 100 100 100
Regulation
Load
Following Down
Forecast
Error Down
, The hypothetical nuclear plant provides 100 MW of energy across the hour in all
operating cases. The plant is base loaded and unable to move within the hour. Coal
on the other hand, has some modest intra-hour flexibility. In Case A the plant
generates 100 MW of energy and 5 MW of down load following. The coal plant has the
ability when generating at capacity to reduce its generation during the la-minute
window by 5 MW; it therefore can provide 5 MW of this service. In Case A both the
nuclear plant and the coal plant are obtaining their maximum value in the energy
marketplace.
In Case B the coal plant is dispatched to meet both 5 MW of down load following and 5
MW of up load following to cover variability in both directions. In order to provide the
f n e e'XCORPORATION Page 22
capability to follow load in the upward direction, it is necessary to schedule the plant to
produce 95 MW of energy.
Finally, Case C illustrates that even where the coal-fired plant is schedu1ed'at the lower
level of 30 MW, it still is' able to provide only 30 MW each of down and up forecast error.
The limitation in this case is not the capacity of the plant, but that it can only move 30
MW in any single direction during an hour (5 MW x 6 10-minute intervals) load. This
later case likely would describe a condition whereby the coal plant was being operated
at a loss in the energy marketplace and went into the hour running at a low level to
minimize this loss while still providing reserve capabilities.
A hydro plant has very low operating costs and can ramp all of its capacity to meet
various reserve products when called on. In Case A the plant is run to produce 100
MW of energy. Its ability to follow load down is split evenly between down load following
and down forecast error. This scenario likely would occur during peak hours of the day
when market prices are at their highest.
In Case B it is necessary to provide both up and down regulation and up and down load
following. The hydro plant lowers its energy production so that it can provide these
additional services. This operation profile could be a very expensive one from the
perspective of energy production. During a high demand hour market prices for energy
could be very high, meaning the utility is losing the opportunity to sell 50 MW of energy.
Where market prices are low, the hydro plant might be running to make available
reserve products and losing a lot of money relative to where it was not required to run
at all.
Case C provides a third look at hydro operations , but shows how a plant can be used to
serve load regulation and load following, as well as forecast error, simultaneously.
f n e e'XCORPORATION Page 23
Section 5
MODELING AND ASSUMPTIONS
OVERVIEW OF THE AVISTA SYSTEM INTEGRATION LP MODEL
The Avista LP Model represents a true system dispatch of Avista generation and
contract resources against its control area loads. All resources and obligations are
modeled hourly in one-month time steps across many months. Hydro project storage is
modeled to minimize system costs over time while reflecting operational and
environmental obligations. Hourly deficiencies and surpluses are balanced in the
wholesale market, limited by transmission availability.
The LP Model was developed in Microsoft Excel. A linear programming add-in to
Microsoft Excel What's Best! by Lindo Systems, is used to optimize operations in all
cases. Four system optimization modules in the LP Model represent four unique areas.
The first two represent two hydroelectric storage projects. Each of these modules
optimizes a hydroelectric project to maximize its value given constraints in the
remaining modules. The third optimization module represents the pre-schedule
time frame where forecasted resource availability is scheduled on a day-ahead basis
against forecasted obligations. The last optimization module is similar to the pre-
schedule LP Model, except that it replaces forecasted data with actual data.
Data used by the LP Model are contained in a Microsoft Access database. All results
are stored in another Microsoft Access database. A schematic of the pre-schedule
module is shown in Figure 17.
t n eCORPORATION Page
Figure 17: Schematic of Avista LP program - Pre-schedule module
The Controls Module
The Controls module (Figure 18) sets up various modeling runs. Start Date for the
run is specified , and must begin on the first day and hour of a month within the study
horizon. As explained earlier, the LP Model runs hourly in one-month time steps. The
Months field can be adjusted to instruct the LP Model on how many contiguous months
are to be run. The Run Description provides a unique identifier of the run so that it may
be found within the output database.
The next three lines allow the user to specify how many megawatts (MW) of nameplate
wind capacity will be integrated at three available locations. The three locations can be
mixed and matched" to provide system diversity.
f n e e'xCORPORATION Page 25
Start Date
Months
600,
11.
20%
15,
AVGI
$0.
$4,
$0.
$0,
$1'
801)
823)
Henry Hub Basis Differential
What'sBestR,esult (WBMIN)
Res~ltant Costs ($millions) , "
WiiteOLJput Results to Database
Input Database Table Name
, ,
OutpuLDatabas~Name (wtexl)'
Master Table New LF 2
MCWIND AVGWTRmdb
$0.
$65,
$3.
' '.. ":,::/::,
spiIlPen~ltYWMVYh)'
. "
ActualWind Feather Pe~~IW' ($tIy1Wh) "
Additiol1alTransmissionpurchase?
, "
AdditionaITra~sn'issiQn ~s(($tMWh) ,
.. ,
Figure 18: A vista LP Model control interface
Load regulation, load following, and wind forecast error are the three incremental
reserve products the LP Model evaluates to determine integration costs. Incremental
Wind Regulation (MW) provides a placeholder for additional regulation requirements
necessary to integrate incremental wind resources. Wind Forecast Error is represented
as a percentage of installed nameplate wind capacity. It allows the user to vary the
perceived accuracy of the hour-ahead wind load forecast and determine the impact.
The Forecast Error Credit offsets by the specified amount the incremental wind forecast
error obligation in each hour. This value reflects the forecast error of the Avista system
for both load and generation variability on the hour-ahead timeframe.
The Controls module allows the user to specify whether a wind delivery is "Flat" or not.
Today s generally accepted wind integration analyses method is to consider wind on a
system , and then in the comparison case deliver a flat (i.e., across all hours of the
analysis period) quantity of wind that is equal on an energy basis.
Wind integration costs on a hydroelectric-based system vary depending on how much
water is in the system. The Model is capable of running low, average, and high water
year scenarios. The Market Price Differential, Spokane Price Differential, Clark Fork Spin
Penalty, and Cost of Reserve fields generally are not changed by the user. These fields
provide incentives for the LP Model to avoid certain behavior. For example, it is not
efficient, or realistic , for the LP Model to purchase and sell thousands of megawatts of
power in any given hour. Unless there is some price differential, the LP Model will
randomly buy and sell too much power. Inserting a modest differential between the
f n eCORPORATION Page 26
market price for sales and purchases solves this logic error without affecting overall
results. A similar incentive is used to prevent the LP Model leaning on the Clark Fork
hydroelectric facility for spinning reserves, and ensuring the LP Model provides only
those reserves necessary in any given hour to meet load and wind obligations.
The LP Model uses Henry Hub for its natural gas price history; the Henry Hub Basis
Differential was estimated to be $1.00 per decatherm. The next two fields explain the
financial results of the optimization routine. The What'sBest Result (WBMin) field
provides the actual solution of the LP routine. The Resultant Costs ($millions) field
provides the actual value used in the wind integration calculation. This field ignores
artificial adders and penalties used to help the LP Model to emulate certain behaviors
(e., not to over-provide operating reserves), as described in a previous paragraph.
The next three fields tell the Model to or not to write its results to a database, what the
input database name for the run is, and the name of the database where results are to
be written to.
The Wind Feather Penalty incents the LP Model to not feather, or spilling, generation.
Under rare certain circumstances Avista s system cannot integrate all wind generation
at high wind penetration levels; wind must be feathered for the LP Model to solve.
When wind is feathered, the project owner looses the federal production tax credit. This
value is approximately $20 per MWh above the wholesale cost of power. Unlike many of
the penalties discussed in this section, the Wind Feather Penalty is carried through to
the ultimate solution and adds to integration costs.
The LP Model allows the user to specify if additional firm transmission was purchased
or constructed for the project. The Additional Transmission Purchase? field ultimately
affects how much energy can be sent to or delivered from Avista s system; this affects
wind integration costs. For this study finn transmission purchases were assumed for
the full nameplate capability of all added wind generation.
Short-term transmission purchases are made where long-term contracts are incapable
of meeting model requirements. It is assumed to be purchased from the Bonneville
Power Administration at a cost of $3 per MWh, excluding losses. Short-term
transmission purchases in any given hour are limited to 300 MW (total of 540 MW
including firm transmission rights) in the east-to-west direction and 760 MW (total of
000 MW including firm transmission rights) in the west-to-east direction.
The Assumptions Module
The Assumptions module details various operating characteristics and capabilities of
Avista s portfolio , including operating reserves and transmission losses. Firm and non-
firm transmission availability are also represented. Input for the Assumptions module
is shown in Table 13 and Table 14.
f n e e'XCORPORATION Page 27
Table 13:Avista LP Model Resource Assumptions (two tables)
RESOURCE
ASSUMPTIONS
Capacity... _no-----
Heat -Raie
-~'---
yrn;;s"NOde
---- -"'-
Nori~sPiiiRes
:------
spiiiriiiig-Res
-'-
vartiiEfe O&M
---
siiin'ResourCe
----'-'-
rifmum C3en
--- -"'--:~"--
siora9'e
, HIK Factor:
Spokane
Ri~r
180
HYDRO RESOURCESNoxon CabinetRapids Gorge
554
Yes
006
3.788
236
Yes
933
DISPATCHABLE GAS RESOURCES. Mid Coyote Boulder Rathdrum Rathdrum
Columbia Springs 2 Park
138..4
Yes
569
280
100
SP/MC
180
000 12.000 000
DISPATCHABLE GAS RESOURCES FIXED RESOURCES
RESOURCE . Kettle . NOrtheast
, ,
rthe~st Colstrip .' Kettle' MC Fixed SA Fixed
ASSUMPTIONS . FallsCT" 'A "." B
. ,
Falls " Contracts Contracts
. '' .
Capacity
----
~eai-Riiite
'------:--
TransNOde
-----._,-----_. -..-..-~._--
~'?!t~pi~_
~~~,--_._....~~,:!~,!;!_
~es
Variable O&M
~:,-
J~piE~e~2~~~
----,.
- Mi!1!.'!1.rI!_
" "
Storage
HI~ Factqrj
750 000 13,000
Table 14:Avista LP Model Transmission Assumptions
. Reser\e Requiremt
-:--
i=iiTI1Transii1Tssid
C'-~ihjnSLOsseS:
' .
' Ma~-~!~ns
' ..
MaxW2E;Trans
NODES
~pokane
- Node .,
240
540
000)
222 125
None
None
125
None
None
The Clark Fork logic and Mid-C logic Modules
The Clark Fork Logic Mid-C Logic modules dispatch Avista s hydroelectric projects that
have intra-week energy storage. The modules take daily average inflows into each of the
projects and optimize generation and spill levels over time. The results are handed off
to the Pre-Schedule and Real-Time modules for within-day optimization. The Mid-
Logic module is illustrated in Figure 19.
f n e e'xCORPORATION Page 28
NOXON DAILY GENERATION AND SPILL
; .
Project Noxon Begin
Run Inflow Inflow Storage PSStorage PSStorage PSStorpge PSSlorage RIDaily RIDaily RTEnd
Description Date (sid)(MWh)(MWh)Max Con Min ,~,'"Change Change Gen Spill Storage
6COMW MC AVG O,2%FCS'toRR 7/1/2003 500 068 806
=,=
12,266 605 003
600MW MC AVG 0.2%FCSTERR 7/212003 46,400 12,250 003 12,416 837
600MW MC AVG 0.2%FCSTERR 7/312003 48,100 698 837 546 989
600MW MC AVG O,2%FCSTERR 7/4/2003 800 13,411 989 12,542 792
600MW MC AVG O,2%FCSTERR 7/5/2003 600 566 792
=~=
12,352 006
600MW MC AVG 0.2%FCSTERR 7/6/2003 700 11cDQ9 006 11,178 837
600MW MC AVG 0.2%FCSTERR 7/7/2003 37,700 953 837
~=
984 806
600MW MC AVG O.2%FCSTERR 7/8/2003 100 266 806 454 618
600MWMC AVG 0.2%FCSTERR 7/9/2003 800 187 618
=~-
799 006
600MW MC AVG 0.2%FCSTERR 7/10/2003 900 8,422 006 762 666
600MW MC AVG 0.2%FCSTERR 7/11/2003 27,500 260 666 063 864
~~MW MC AVG 0.2%FCSTERR 7/1212003 33,600 870 864 339 395
600MW MC AVG O,2%FCSTERR 7/13/2003 100 002 395 391 006
600MW MC AVG 0.2%FCSTERR 7/14/2003 000 184 006 385 806
600MW MC AVG 0.2%FCSTERR 7/15/2003 32,400 554 1,806
"'-
353 006
600MW MC AVG 0.2%FCSTERR 7/16/2003 200 501 006 872 635
600MW MC AVG O,2%FCSTERR 7/17/2003 31,400 290 635 557 367
600MW MC AVG 0.2%FCSTERR 7/18/2003 000 8,448 367 872 943
600MW MC AVG 0.2%FCSTERR 7/19/2003 500 844 943 192 595
600MW MC AVG 0.2%FCSTERR-7/20/2003 31,400 290 595 721 163
600MW'MC AVG 0.2%FCSTERR 7/21/2003 23.300 151 163 509 806
600MW MC AVG 0.2%FCSTERR 7/2212003 24,400 442 806 044 204
600MW MC AVG 0.2%FCSTERR 7/23/2003 500 6,468 204 665 006
600MW MC AVG 0.2%FCSTERR 7/24/2003 14,000 696 006
=,=
499 204
600MW MC AVG 0.2%FCSTERR 7/25/2003 500 092 204 895 401
600MW MC AVG 0.2%FCSTERR 7/26/2003 23.900 310 401 074 637
600MW MC AVG 0.2%FCSTERR 7/27/2003 500 148 637 4,495 290
600MW MC AVG 0.2%FCSTERR 7/28/2003 300 359 290 843 806
Figure 19: Mid-C Logic Module
The Pre-Schedule Model and Real-Time Model Modules
The Pre-Schedule and Real-Time modules are too large for a visual representation in
this discussion. Both are similar in organization, with the significant difference being
that the Pre-Schedule module uses forecasts of load, market prices, and wind
generation in its optimization routine. The Real-Time Model uses actual values for
these variables. Each module perfonns an hourly optimization of resources against
loads , honoring resource constraints , transmission paths, tracking reserve obligations
and balancing the portfolio using the wholesale marketplaces for natural gas and
electricity. Purchases and sales entered into during the pre-schedule time frame are
carried through to the real-time as obligations that must be met in addition to real-time
loads and resources.
SCENARIO ANALYSIS AND OBJECTIVES
The exact impacts of future wind acquisition are uncertain. It will be many years before
we learn how accurate our forecasts of integration Gosts are. Many studies have
provided integration cost estimates based on a set of assumptions. Some scenarios
were studied to detennine the sensitivity of wind integration costs to various key
changes in assumptions. More scenarios are necessary.
Limitations in scenario analysis oftentimes are the result of long solution times
necessary to model systems accurately. Wind integration studies require a complex
level of analysis beyond traditional engineering-economics studies. Production cost
models oftentimes have been used to determine how a larger system changes its
dispatch in response to bringing wind online.
This study benefits from a new LP Model developed internally by Avista over the past six
years. The LP Model focuses on the re-dispatch of Avista resources balanced by an
hourly wholesale electricity marketplace. Instead of modeling all generation resources
f n e e'XCORPORATION Page 29
in the Western Interconnect, hourly market prices represent the world outside of
Avista s control area.
Solution times for the Avista System Integration LP Model are short , requiring
approximately thirty minutes for a one-year control area dispatch. This efficiency lends
itself well to more scenario analysis.
Debates continue in our industry about the level of incremental regulation, load
following, and forecast error reserves necessary to integrate wind. This study, while
identifying a level of integration cost, focuses on quantifying sensitivities around these
levels. Many in the wind industry advocate better forecasting to reduce wind
integration costs. Irrespective of the ability of the forecast industry to provide better
forecasts , it is difficult to decide whether or not to pursue such avenues absent a means
to measure what a better forecast would mean for wind integration costs. By studying
varying levels of forecast error, one can determine the value of a better wind forecast
and spend resources appropriately.
Scenario analysis for this study falls into five categories. The first explores wind
diversity to learn how much costs potentially change where a geographically diversified
mix of wind farms is pursued. The second looks at system wind penetration levels to
evaluate how integration costs change as wind become a larger share of the utility
generation mix. Third, as the Northwest marketplace is driven by hydroelectric
demand, the impact of varying water conditions is explored. The fourth scenario
reviewed the impacts of market price levels on integration costs. Finally, as forecast
error levels were found to represent a significant portion of the total cost of wind
integration, and in fact are a large portion of the debate around wind integration today,
varying levels of wind forecast error were modeled.
This study considered four mixes of wind generation: a mix of Columbia Basin wind
farms, a mix of wind farms in Montana east of the Continental Divide, a combination of
Columbia Basin and Montana wind farms and a diversified mix of five wind sites located
across the Northwest and into eastern Montana. The impacts of these mixes are driven
primarily by their reserve obligations. Montana wind tends to be more volatile
requiring more reserves than Columbia Basin wind. The diversified mix provides the
lowest mix of incremental reserve requirements, driving it to have the lowest integration
costs. Table 15 details the estimated reserve requirements of these wind resource
mixes.
Table 15:Total Operating Reserve Assumptions for Wind Scenarios
. '. '
' Forecast'In " ysem
, ,
.' In Regulation load Follow .
." " .
Total
' .
, %of
, ", '" "' .
ErrortCapadtyPenetration ., Location, .((V\Wh (MWy "
." .
".' MW " (MWr . amepa e
100MW C. Basin 1.3
200 MW 10%50/50 Mix 14.
400 MW 20%Diversified 15.15.38.
600 MW 30%Diversified 11.0 27.7 30.68.11.
OPERATION OF A HYDRO-BASED CONTROL AREA
Avista, like the majority of its Northwest peers, operates a control area backed
predominately by hydroelectric resources. Hydroelectric generation plants offer
tremendous flexibility when compared to other generation technologies. In many cases
~ n e e'XCORPORATION Page 30
a turbine can be ramped from zero to full capacity almost instantaneously. This
flexibility makes hydroelectric turbines ideal for covering variations in generation and
load, including variation due to wind generation. A second characteristic of
hydroelectric generators is much like wind generation plants: zero fuel cost.
Hydroelectric generation plants generally are "energy-limited " in that they do not have
enough fuel to operate during all hours at maximum capacity; they have vast capacity
relative to their energy generating potential. This is in contrast to other traditional
resources that have essentially unlimited fuel supplies , but instead are limited by their
generating capacity. Hydroelectric facilities tend to be operated over peak and super-
peak hours of the day to maximize the value of their limited energy generation potential.
Water is stored overnight and, where adequate storage exists behind a dam, on days
with lower demand and market prices. This stored energy is then shifted to higher
value periods.
Maximizing a hydroelectric facility s value is affected by the level of reserves necessary
to balance control area loads and resources. Higher reserve levels generally necessitate
hydroelectric generators operating away from their optimal energy generation points.
Energy not generated during peak hours must be shifted to less valuable shoulder
hours. Figure 20 provides a graphical depiction of this impact.
250
No Reserves $22 500
100 MW Reserves $19 000
200 -
~----------- ---
---______H___--_--- 40
S: 150
c::
~ 100 -
c::
30 ~
C,,)
;::----
20'::-
C1I
--------. ---.-- ,_.
-- 10
Hr A-Oft-Peak Hr B-Peak Hr C-SprPeak Hr D-Peak Hr E-Off-Peak
period of day
- No Reserves E::'I100 MW Reserves -Market Price
Figure 20: Maximizing Hydro Facility Value
The figure shows how a 250 MW hydroelectric plant would theoretically dispatch
against a set of five market prices representing a 5-hour day. Total inflow for the day is
500 MWh. In the No Reserves case , generation is focused on the two highest-value
hours, generating 250 MWh in each period-for a total value of $22 500. Where 100
of reserve are carried in the second case , the maximum generation level in any hour is
2 In some cases, changes in thermal plant operations can be made at a lower cost; but, generally it is
hydroelectric units that provide reserves on Avista s system,
f n e N e'xCORPORATION Page 3
lowered to 150 MW. This forces the hydroelectric plant to maximize its value by
generating during four hours for a total value of$19 000. The reserve costs in this
example equal $3 500, or $7 per MWh ($3 500 / (100 MW * 5 hours)) ofreserves.
Economists define this as opportunity cost. Providing reserve capacity de-optimizes
generation efficiency on a hydroelectric system. Wind integration magnifies reserve
obligations, thereby increasing opportunity costs. This is the key concept underlying
this study.
f n e e'XCORPORATION Page 32
Section 6
IMPACTS OF WIND GENERATION WITHIN THE HOUR
MODELING AND ANALYSIS FOR WIND INTEGRATION ASSESSMENT
The common methodology for assessing the cost of integrating wind energy into a utility
control area is based on chronological simulations of scheduling and real-time
operations. Production costing and other optimization tools are generally used to
conduct these simulations. In most cases , the "time-step" for these simulations is in
one-hour increments. Consequently, many details of real-time operation cannot be
simulated explicitly. Generation capacity that is used by operators to manage the
system in real-time - i.e. the units on AGC utilized by the EMS for both fast response to
ACE and that which is frequently economically re-dispatched to follow changes in
control area demand - is assigned to one or more reserve categories available in the
various programs.
At this level of granularity, the total reserve requirements for the system are a
constraint on the optimization and dispatch. Supply resources are designated by their
ability to contribute to system requirements in one or more reserve categories. In the
course of the optimization or dispatch, the solution algorithm must honor system
reserve needs, and therefore is not able to use some capacity to meet load or fulfill
transactions.
In this context, there are two primary types of reserves. The first is comprised of the
excess capacity that must be carried at all times for reliability. These are generally
known as "contingency reserves , and as the name implies, can only be utilized when a
contingency actually occurs.
The second category of reserves is used to balance the supply with the control area
demand on a continuous basis. This includes minute-by-minute (or faster)
adjustments to generation to compensate for load variations and frequency economic
dispatch of units with movement capability to follow slower variations in control area
demand.
There are periods where demand is higher or lower than the average over an hour.
Generation must be adjusted to meet these values within the hour. Figure
illustrates this with actual data.
~ n e e'XCORPORATION Page 33
1550
Scheduling and dispatch
simulations are based on
the hourly average value
14 9 O--
~~-~ ~.~ ~.~~~ ~~.!~.~~~~~_...:
1470 m Hourly minimum load
--
- Within the hour
, generation must
. move over this
....... -
range
:::E
1450
2991 2991.2991.5 2991.75 2992 2992.2992.2992.75 2993
Hour
Average Hourly Load
Load (10 min. resolution)
Figure 21: Hourly average and ten-minute load
CALCULATING REQUIREMENTS FOR MANAGING VARIABILITY WITHIN THE HOUR
The purpose of this study is to develop a procedure for estimating the additional
flexibility within the hour that would be required to manage a control area with
significant wind generation. The analysis and experimentation are based on an annual
record of load and wind generation at ten-minute intervals. The goal is to develop a
rule" for the amount of flexibility that would be required using information that would
be available in the control room. The extended data records also provide a way to test
the proposed rules.
The procedure for determining the required flexibility for load alone is as follows:
1. Using the ten-minute data, compute the hourly average value for load
2. Compute the difference between each ten-minute value ofload and the hourly
average. The difference is the load following requirement.
3. Because of defined WECC ramp which takes place from 10 minutes before until
10 minutes after the hour, the average load value at the top of each hour is
actually the average of the previous and next hour values (Figure 22). This
adjustment will reduce the magnitude of the hourly load following "envelope
since the greatest departure of ten-minute values usually occur at the start and
end of each hour using this method.
4. Devise an algorithm that could be implemented by operators to project the
maneuverability needed to follow the load movements. For load alone , this
algorithm is based on the previous hour average value (which is known) and
the forecast average value for the next hour (which we will.assume can be
perfectly forecasted for load alone).
5. The estimated load following capability is then the difference between the next
hour forecast average and the previous hour average.
t n e e'XCORPORATION Page 34
6. The requirements are roughly symmetrical about the average value. In the
morning, for example, the load at the beginning of the hour will be less than
the hourly average. If the unit base points are moved to the hourly average
there will be a need to back some generation down, and then move it up over
the hour as the load increases.
7. This load following rule is tested with the ten-minute data. The number of ten-
minute load values outside of the up and down load following bands is
computed. For the rule above, the number of "violations" is about 1 800 out of
almost 50 000 ten-minute samples. It was also necessary to add in regulation
capacity, as the ten-minute values are snapshots, not interval averages.
Figure 22 and Figure 23 show the results of the mathematical procedure described
above in points 1 and 2.
1550
1530
1510
1490 .
1470
1450
2991
+-10 min-++-lO min+
,..._-----.............-.,,,-...-----.
2991.25 2991.5 2991.75 2992 2992.2992.2992..75 2993
Hour
Average Hourly Load
Load (10 min. resolution)
Figure 22: Hourly average and ten minute values. with over-the-hour ramp period
( n e e'XCORPORATION Page 35
16C'~
1120 m..
~~=_
0 --'i--
--
_n_--
I--- 0
- --
1-- .----0 ,1040----
---'-' --- ~ __
-1_.n.___
+_-- ---
, -r-,._---t------- +-- o_..o...,-
---
; i
::~ ~ =~ - ~~~==-~-
l ~ = --~= =I==~=-~:= C~=~, - --1=::
~ =~~ -~=,
c==~~ ~=~- -- ~
~ ~-== -
==L
Beo
64CO 6410 6420 642,0 6440 645-0 646.0 6470
Hour
Average Hourly Load
Load (to-min.
SLF Up + LlO
SLF On - LlO
Figure 24: Ten-minute average load shown with up/down intra-hourly load following capability
From this baseline, incremental reserves are added in each wind case to maintain the
same level of CPS2 compliance. Both wind generation and load are assumed to be
forecast perfectly, so the hourly average value from which the deviations are computed
is the net of the hourly average load and the hourly average wind. The amount of wind
generation change over an hour is the metric for characterizing wind generation
variability. There are other metrics that could be developed, but this study s approach
lends itself well to the data.
Using hourly average wind generation data, variability over one hour is computed for
ten deci1es of production. The results for the Avista Mid-C wind scenarios are shown
in Figure 25. The curves show that the maximum variability occurs in the mid-range
of aggregate wind capacity.
t n e e'X
CORPORATION Page 37
100
100 MW
200 MW
400 MW
600 MW
-'C
. "'- '
0.46
Production Level (pu)
Figure 25: Variability of wind generation over one hour from LP Model data (by scenario)
The empirical results from Figure 25 are approximated as quadratic expressions , with
the input to the expression being the current average production. This facilitates a rule
that can be applied on an hourly basis. In this first example, reserve planning for the
hour is performed just prior to the start of the hour, so that the average production is
from hour t-l and the amount of change predicted for hour t.
100
-'C
-lOOMW
-- --- 200 M
. 400 MW
- , - 600
---- -' -- -- -....::-
c:-
120 180 240 300 360 420 480 540 600
Production Level (MW)
Figure 26:Approximation of empirical wind generation variability with quadratic expressions
The quadratic expressions are:
(x- 60)f 1 (x) := 1 4 -
300
~ n e e'XCORPORATION Page 38
(x- 110)f2(x) := 20-
1700
(x - 300)f4(x):=32-3500
(x-475)f6(x):=50-
6000
where f1 through f6 correspond to the 100 MW, 200 MW, 400 MW, and 600 MW
scenarios respectively.
The load following rule for each wind scenario is of the form
(HWind1 hI-I - 60)2F1hl:=FOhl+k1.15-
300
where the variable in the expression is the current hour average wind generation (h-
because we are planning for hour h), the quadratic constants are from the empirical
analysis described previously, and FO is the load following requirement for load alone.
The coefficient k1 is adjusted so that the number of CPS2 "violations" is the same as for
the case with no wind-about 1 500 with a 20 MW band of regulation capability.
Running these experiments for each wind generation scenario, the following coefficients
are determined:
k1 = 0.
k2 = 0.
k4 = 0.
k6 = 0.40
The rf;':quired additional load following capability is much less than one standard
deviation of the hourly change for all cases. Also, the coefficients will vary depending
on the nature of the wind generation scenario. Concentrated and correlated wind
generation facilities would lead to higher coefficients, while well-distributed scenarios
would tend to reduce them. The scenarios developed for the Avista study bear this out.
Figure 27 depicts the hourly load following bands for the 400 MW Mid-C wind scenario
for the same three day period shown in Figure 24.
f n eCORPORATION Page 39
16':0
1120
H__n n
i'-'-- ---
1':,:' --
...-.
6+:0 64l-J 64~:'642.6440 64E,J 640.0 S47.J
Hour
Average Hourly Load
Load (lo-min.
SLF Up + LIO
SLF On -- LIO
Figure 27: Ten-minute load net wind generation and intra-hourly load following capability
The average load following capabilities over all hours of the sample year for the four
wind generation scenarios are shown in Table 16.
Table 16:Average Hourly Flexibility Requirements for Managing Control Area Variability
,..,
C'aseo .i'\
)/...'
:AverageHourly Flexibility. (+l~r
' "," '" '" "" ,. ,
Load only 20MW
100MW 22.3 MW
200 26.6 MW
400 34.8 MW
600 44.6 MW
IMPACTS OF SHORT-TERM FORECAST ERROR ON REAL-TIME OPERATIONS
The previous analysis assumes that the reserves for the hour are planned on the basis
of perfect knowledge of the next hour average load and wind generation. This is the
situation with the minimum uncertainty, and relates mostly to the real-time operation
of the system to compensate for inside~the-hour variations from some constant average
value. In reality, there are operational decisions made some hours prior to this hour
that will affect the generation flexibility that is needed to manage the control area.
If reserves must be allocated an hour or more before the operating hour, the known
wind generation at that time may be substantially different than in the hour in
f n e e'xCORPORATION Page
question. This could impact the projected variability, as it is a function of the current
production level. However, since the variability curves (Figure 25 and Figure 26) do not
change dramatically with slight changes in production level, the error here would be
slight.
Larger impacts stem from decisions made based on short-term forecast information. If
the window for hourly transactions closes one hour prior to the hour, it is necessary to
cover deviations (i.e. forecast error) in the average hourly load net wind from the
forecast hourly average load net wind (Figure 28). These deviations are covered by
internal generation capacity which has been set aside for the hour in question since
there is no other alternative. The deviation is constant through the hour in question
and is actually an offset in the operating position (Figure 29). To cover the deviation, a
resource must be scheduled at an operating point for the hour different than what was
planned when setting up the hourly schedules. This action is not really following the
load , but rather addressing a energy deficit or surplus from schedule. Generation
capacity must be reserved to make this adjustment.
1500
Actual Hourly Average
Forecast (g) H-
(j)(j)
1400
1300
3552 3554 3556
Hour
Figure 28: Actual and forecast hourly average values. Short-term forecast is made 1.5 hours prior
to the start of the subject hour.
f n e e'XCORPORATION Page 41
1:1
,..
IJ)
IJ)
1450
Required "" flexibility due to,r--- variability and schedule error
iOCC
1400
Flexibility required for variability only
~schedule (i.e. forecast) error
Actual Load
Hourly Scedule
variability wo/ Schedule Error
3551 3551.5 3552 3~2.35S3
Hour
Figure 29: Additional intra-hour flexibility requirements due to schedule error bias.
Schedule deviations are a consequence of short-term load and wind generation forecast
errors. Avista currently carries approximately a I5MW band to cover load variation.
The schedule deviation will be larger with wind generation. An approach similar to that
used to calculate incremental regulation and load following reserves can be employed to
determine how much additional capacity must be allocated to cover incremental
forecast error. The error in a persistence forecast over a two hour horizon is calculated
from the hourly wind generation data and summarized in Figure 30. Note that the
standard deviations here are larger than for the I-hour persistence forecast (which
would correspond to Figure 25 and Figure 26), illustrating the relatively rapid
degradation of the persistence assumption over longer time frames.
f n e e'XCORPORATION Page 42
100
lOOMW
----- 200
-- 400
- - - 600
"t;
"t;
ill
a....
:r:
-- -- -- -- '.......- '----' -;;.~:
r'\
120 180 240 300 360 420 480 540 600
Production Level (MW)
Figure 30: Standard deviation of persistence forecast error over a two hour horizon for the four
wind generation scenarios.
Quadratic formulas for the curves of Figure 30 were added to the equations for hourlyreserves, and the coefficients adjusted to achieve the same control performance as for
load variability alone, Load forecast error was modeled as a normally distributed
random variable with a standard deviation of 7.5 MW (one-half of the 15 MW Avista
currently carries to account for short-term load forecast error). Table 17 shows theresults.
Table 17:Total Reserves for Variability and Schedule Deviations
."'' ."" ", "
verag HourlyFlexib
, "
verage our y exi II
" "
Cas
ty,
rVanabllttyand
. '
' or anall
, ,, . ", , " ''. "' .,' , ",' ", ' ",. '
Schedule.Devlatlon,
(+/~)
Load only 20MW 35,0 MW
100MW 22.1 MW 38,3 MW
200 MW 24,1 MW 49.5 MW
400 MW 27,9 MW 68.7 MW
600 MW 31.0 MW 103.7 MW
Where both load and wind generation forecast errors are random variables, the
schedule deviation error planned in advance of the operating hour would be the root-
mean-square value of the respective standard deviations. With 100 MW of wind
f n e e'XCORPORATION Page 43
generation , the component of reserves is increased from 15 MW for load alone to 32.
MW. This incremental amount is very close to the root-mean-square value of standard
deviation ofload (7.5 MW) and the two-hour persistence forecast error for 100 MW wind
generation scenario (14.6 MW). This relationship holds for the other scenarios , with
wind generation forecast error becoming the dominant factor at the higher penetration
levels.
Just prior to the operating hour, the direction of the forecast error will be known. Intra-
hour variability, as computed earlier in the study, must still be covered and is not
affected by the forecast error. So, it seems that the real-time operators would know at
the beginning of the operator hour how the scheduling error would impact the reserve
requirements. If there is additional energy to be provided to cover the forecast error, the
capacity set aside to move up would be used, with no need to retain the downward
movement capability. The operating plan for the hour must be sufficient to cover both
the up and down side of the forecast error.
Because it is an offset in the flat schedule for the hour, there is minimal intermingling
with load following error. While the same resource may in fact be called upon to
address both, computation of the individual requirements in developing the plan for the
hour is separate.
f n e e'XCORPORATION Page 44
Section 7
RESU L TS
BASE CASE RESULTS
Total integration costs for the four base scenarios are detailed in Table 18. Costs range
from $2.75jMWh of wind generation for the 100 MW Columbia Basin scenario to
$8.84jMWh for the 600 MW diversified mix, which equates to a.30% capacity
penetration level.
Table 18:Integration Costs for Base Scenarios
. .
Wind .
' .
Wind"
' .
System."Forecast Cost 'Cost
. '
.' Locatio"". Capacity Penetration: ,
' .
' Error ' . ($/MWhY (% Mktl
Columbia Basin 100MW 15%$2.
SO/50 Mix of CB & MT 200 MW 10%10%$6.12.7%
Diversified Mix 400 MW 20%$6,12.
Diversified Mix 600 MW 30%$8.16.
The incremental reserve requirements are detailed in Table 19. The ratio of the
incremental reserve amounts for the larger penetration scenarios is consistent with
what has been reported in numerous studies. The costs are also in the range of what
has been reported in previous North American wind integration studies.
Table 19:Incremental Reserve Assumptions for Base Scenarios
, W
.."
System ..Wind LJlotion.Load Foilow,Forecast'..c tal %of
. ", "' ". " '' .
. c'
. '
. Error,Capacity Penetration. Location " (MWL '
'.' , ,,' .
' (MW)
/ .' .
' , MW' .
. .
.. (MWY Nameplate.
100MW C. Basin 1.3
200 MW 10%50/50 Mix 14.
400 MW 20%Diversified 15.15.38.7
600 MW 30%Diversified 11.0 27.7 30.68.7 11.
By changing certain input assumptions, it was possible to determine the contribution of
various factors to integration cost. Table 20 and Table 21 decompose the integration
costs calculated for the base cases into four components. The "wind shape" cost is the
monetized difference of the market value of the actual wind delivery relative to the proxy
resource shape. Regulation and load following are attributable to the opportupity cost
of the incremental reserve capacity required to manage the additional variability of the
control area demand with wind generation. The Forecast Error component relates to
f n e N e'XCORPORATION Page 45
the additional capacity that must be reserved to cover deviations in actual wind energy
delivery from the short-term (hour + ahead) forecast.
Table 20:Components of Wind Integration Cost - Dollars
\ Wind Sys~em-ii~ind Wind Reg Load Forecast Total
,c:apafity Penetrationiil..?c:ation Shape iulation , Following i' '. Error
' '
, Cost ,
".'"""."'
. $/MVvh'i $/MWh .,$/MWh ' $!MWh $/MWh;
100 MW C. Basin 1.13 1.03
200 MW 10%SO/50 Mix 0.44 1.62 1..70
400 MW 20%Diversified 1.67 1..7 9
600 MW 30%Diversified 1.43
Table 21:Components of Integration Cost - Percent
' ..
Wind;.
" ,
system
i.
winct vvind,..Reg-
i".
, Load
. ,..,
, Forecast
Capac;i
tyi Penetration-
Location ' Shape".' ulation ,, Following ( "" Error)
' ,' ,', ,"', ."
$/MWh " $/MWh"$/MWh", $/MWh'
100MW C. Basin 10.40.37.10.7%
200 MW 10%SO/50 Mix 23.46.24.
400 MW 20%Diversified 25.26.40.
600 MW 30%Diversified 16.43.33.
A significant portion of integration cost stems from changes to hydroelectric operations.
These plants operate less efficiently to provide the incremental reserves necessary to
integrate wind, as shown in Table 22.
Table 22:Hydroelectric Generation Portion of Integration Costs
" ," ,, ", '" ', ' ":",
, Windi.Systern...i , Wi~cf pilled::
...'
Spmed i "Value
, .".:'
' % of.,
" Capacity "Penetration : Location ,. Hydro' '' Hydra", Change ' Integration
.' .. .", '.:",. (%' ', '. "" "
' c'
;,
MWh, "MWh " % OOOs "eercent ,
100 M W c.Basin 3,423 312.42.
200 MW 10%SO/50 Mix 12,818 1,421.0 38~3%
400 MW 20%Diversified 25,952 0.7%630.37,
600 MW 30%Diversified 50.919 1.4%369.4 38.
Hydro conditions affect integration costs. Lower hydro conditions appear to increase
integration costs relative to average and high hydro conditions. Table 23 provides the
integration cost estimates associated with low, average, and high water years. This
~ n e e'XCORPORATION Page 46
result was a bit surprising and further analysis will be necessary to understand exactly
what factors are driving this result. For example, are higher costs driven not by actual
water conditions but by the higher market prices witnessed during a low water year?
Table 23:Impact of Hydro Conditions on Integration Cost
..WIrlct/'
, ./... ... ... .
.Syste r11it:. /
........ :', '
Vvi8~' ..
..~ ;.; ....
Ave rager
..;.
L()0f ;
..
Av~ r age:'...Hi 9 rr..
'Co.oci.. Penefratiori.Locatidn.
...
3'Years.,Hdrd ',.. H drd,Hdro
100 M W CBasin $ 2.$ 3.49
200 MW 10%50/50 Mix $ 8.76 $ 6.
400 MW 20%Diversified $ 9.5.79 $ 4.
600 MW 30%Diversified $ 12.7.80 $ 6.75
Average Market Price $ 54.$62.$ 56.$45.45
SENSITIVITY ANALYSIS
The efficiency of the Avista LP model allowed the execution of a number of additional
cases where input assumptions were modified to assess the impact on integration cost.
Six separate areas were investigated:
Impact of market structure , specifically in the hourly trading that is prevalent in
the Pacific Northwest
Value oflimited wind generation curtailment as a system control option
Value of improved wind generation forecasting
Impact of market conditions on integration cost
Integration benefits of geographic distribution of wind generation
Findings from the sensitivity cases for each of these topics are described and discussed
in the following sections.
Impact of Market Structure
The Northwest marketplace transacts on various time steps with the shortest being one
hour. Other areas of the United States and the world run markets that shorten these
time steps to as short as 5 minutes. Shorter-term markets have a number of costs and
benefits relative to the current Northwest system. A significant benefit of moving away
from an hourly marketplace to one that operates on a 5- or 10-minute basis would
the ability to transact more frequently, thereby reducing reserve obligations
substantially. Wind power advocates , as well as some utility operators; interested in
lowering regulating reserve obligations , have encouraged the Northwest to consider
moving to a shorter-term marketplace; however, to date there is a general consensus
that the costs of operating a shorter-term marketplace would outweigh the benefits.
Table 13 explains that forecast error and intra-hour load following account for between
45% and 75% of wind integration costs. To quantify the potential value of a shorter-
term marketplace, Avista analyzed reserve obligations in a 10-minute marketplace. The
10-minute time frame was selected because it represented the most granular data
f n e e'XCORPORATION Page 47
available for this study. Using the methodologies previously described in this report it
was found that forecast error calculated on an N-2 time frame could be reduced by
approximately one-third, and that load following would fall by two-thirds in a la-minute
marketplace.
With reserve obligations adjusted, Avista re-ran its LP Model under average water
conditions for each mix of wind resources identified in the Base Case. A la-minute
market would appear to provide significant savings in the range of between 39% and
62%. Table 24 shows that savings could exceed $6 million per year for Avista at the
600 MW penetration level.
Table 24:Effect on Integration Cost of Short-Term Liquid Markets
. "..,..'
10-Min "10:Mirt"
' ,, '
Wind System Wind "Basei Mkt ,.' Mkt ' lQ.Min '" Annuat
.' Capacity ,. Pen~tratio
~, ,
Location ."
. ,
Cost! 'Saving~ ,Savings , Mkt,Cost , Sav,ings:
;, '
, $/MWh' '.ercent', $/MWh', ." $/MWh' $OOO/r
100MW C. Basin $2.75 61.$1.70 $1.$490
200 MW 10%50/50 Mix $6.60.$4.$2.$2.456
400 MW 20%Diversified $6.38.$2.$4.$2.994
600 MW 30%Diversified $8.40.$3.$5.$6.224
Value of Wind Curtailment
Control area operators balance resources and loads in real-time , on a second-by-second
basis. Plant forced outages, transmission line outages, environmental obligations (e.
flows for fisheries, thermal plant emission limits), and other factors can force operators
to make changes that they otherwise would not make under perfect conditions. Prior to
this study Avista recognized the importance of having some amount of wind generator
control to manage short-term emergency operations. It also expected that under certain
conditions it would be economically advantageous to displace wind generators for
reasons other than pure reliability. All Base Case analyses included the option to
feather wind generation so long as the wind resource owner was compensated for both
the contract power price and the federal production tax credit.
In contract negotiations Avista has pursued wind plant operational flexibility. Wind
developers , especially those offering to sell under traditional power purchase
agreements, where payments are made only where energy is delivered , have not
historically been excited about bringing their plants down except for reliability. Avista
believes that the major barrier to developer acceptance is a compensation mechanism
where wind generation is displaced, especially in the case where such displacement is
for reasons other than system reliability.
This study evaluated the potential for interrupting deliveries from wind farms, both for
system reliability and for system economics. Additional LP Model runs were made
where wind feathering only occurred for system reliability purposes; no economic
dispatch was allowed. Integration costs did rise , however, even when compared to the
base case where developers were compensated both for their lost energy value and the
value of the lost federal production tax credit in the case of interruption. Table 25
shows that integration costs rise by approximately 20% when the control area operator
f n e N eXCORPORATION Page 48
does not have the ability to interrupt wind generation for economic reasons. It also
shows that the amount of wind curtailment to achieve this significant reduction in
integration costs is modest, especially at lower penetration levels.
Table 25:Impact of Limited Wind Curtailment on Integration Cost
:i ~rf19::(F)"'Sy~t~m', Wi nd;'
( '
( fUBC1S~H \) '.L,indf'C:'ost i,yit~" C han g#-
: Cqpacity Penetration-; tC1cati(Jrii,:~'CQst\"C-uftaHn1ent;nC1,
.""' '", (%),;."'.. .,.:";
iii,
;\,....,
/C'"i:."
..'.;,..,..".
...;.."f~(~~h);'
......,
;(~l'
:' ..\;,
iCC'
un-aiITent;,
,.:'..-....,.. .,...," "
100MW Basin $2.0.4%$3.37%
200 MW 10%50/50 Mix $6.$8.49 21%
400 MW 20%Diversified $6,$7.20%
600 MW 30%Diversified $8.1.4%$10.21%
Though wind energy is not feathered for system reliability in these cases , it was
discovered that feathering for system reliability would be necessary where the company
focused all of its development (i., above 10% system penetration) in one basin or wind
farm due to the increased variability associated with a non-diversified wind portfolio.
The Value of Improved Wind Generation Forecasting
Wind generation forecasting errors, as with load , contribute to a sub-optimal power
system operation, creating a need for additional reserve capacity. The influence of both
day-ahead and short-term (one to two hours) wind generation forecast errors was
assessed through sensitivity cases using the Avista LP Model.
Table 26 shows integration costs where wind generation deliveries are known perfectly
for day-ahead system scheduling. In the Base Case scenarios, approximately 35% of
the total wind integration cost can be attributed to day-ahead wind generation
uncertainty.
Table 26:Integration Costs with Perfect Day-Ahead Forecast (no pre-schedule penalty)
. '' "., "" ,,, ", ,, .' "..,, ,' ,
, Rear-
' '" " ,, "..'
,Wind
.:.'
.S'Istem i;.;Wind ::'6as8
" '
.Time
, :"
Portion; "
." ".,
Capacity , Penetration locatiorf Cost
, ' ,
Cost 'of Base Difference
" ", "" "
$/MWh,$/MWh".ercent $OOO/r
100MW Basin 1.76 63.264
200 MW 10%50/50 Mix 65.268
400 MW 20%Diversified 64.532
600 MW 30%Diversified 65.849
In the shorter term, wind generation uncertainty requires additional operating reserves.
The impact was illustrated in the previous discussion of market structure , as the
f n e e'XCORPORATION Page 49
principal impact of c1oser-to-real time markets is the attendant reduction in operating
reserves required to cover schedule deviations. With perfect day-ahead knowledge of
wind generation, and short-term uncertainty covered by real-time markets, the sole
contributor to integration cost is the additional variability that must be managed to
maintain control performance.
Integration Cost Sensitivity to Market Conditions
For hydro systems, energy markets are a critical factor in system economics. To assess
the impact of energy market prices on wind integration cost, two market price
sensitivity cases were constructed. In the low market price scenario, wholesale prices
were reduced by 50% from the Base Case. In the high market price case, prices were
doubled from the Base Case. As expected, integration costs change in accord with the
assumed market prices , though not in a perfectly linear fashion.
Table 27:Market Price Impacts on Integration Cost
Marketi Wind(System Wind Forecast:. .
. , .., .. '
. Cost .
, .', ..' .
, " Savings
. Case /Capac::i '. Penetratiorl Locationi ', Error $000 ' $jMWh
,. "
.ercent
100MW C. Basin 15.181.90 1.32 52%
Low 200 MW 10%50/50 Mix 10,589.62%Market
Prices 400 MW 20%Diversified 872.51 42%
600 MW 30%Diversified 2,404.55%.
100MW C. Basin 15.920.
High 200 MW 10%50/50 Mix 10.5.792.22%Market
Prices 400 MW 20%Diversified 9,489.13%
600 MW 30%Diversified 20.280.10.45 18%
Impact of Reduced Forecast Error
In the previous section on intra-hour impacts and operating reserves, it was shown that
expected errors in short-term wind generation forecasts, one to two hours ahead
translate into an additional reserve requirement due to the lead time associated with
hourly energy markets in the Pacific Northwest. Forecast error, therefore, is a
significant contributor to wind integration cost. It is uncertain at this time what
improvement can be expected over persistence from state-of-the-art wind generation
forecasting techniques. To illustrate how this component affects integration cost, a
series of cases was run with differing assumptions about the expected forecast error
over the time frame from hourly trading deadlines to the subject hour. Results of these
sensitivity cases for the base scenarios are shown in Figure 31. As the expected error
rises beyond a certain level, integration costs increase dramatically for all scenarios.
The inflection point for each scenario corresponds to the level where the effects of wind
generation uncertainty begin to dominate the overall uncertainty. At low wind
penetrations, for example, the uncertainty in MW is lower than the short-term load
forecast error, and therefore does not significantly increase the total short-term
uncertainty.
f n e e'xCORPORATION Page 50
As the graphic shows, there is significant benefit from improvements in short-term wind
generation forecasting given the current structure of the hourly energy markets in the
Pacific Northwest.
35,
30,
25.'"0
20,
::r::
15.
-......
-.-5%(100 MW Columbia Ba"in)
(200 MW E MT & c. Basin)
7:-,-20%(400MW Diw:rsifi"d Mix)
(600 MW Di;'"rsifi"d Mix)
-i1-10%
...-...-30%
...
10,
,:-,,;.......-....'--'"
10,15.20.25,30.35.
Short-Term Wind Generation Forecast Error (one std. deviation, % oi rated)
Figure 31 : Integration cost as function of short-term wind generation forecast error for base
scenarios,
Benefits of Geographical Diversity
This study supports earlier work indicating that geographical diversity is one of the keys
to lowering wind integration costs. The Base Case wind integration cost curve does not
rise substantially due to the assumption that Avista over time will acquire a
geographically-diverse mix of wind resources. One of the interesting results of this
study is that wind integration costs actually fall modestly when going from a 10% to a
20% wind penetration level. This is not a data anomaly, but the result of moving from a
basin to a S-basin mix of wind projects.
Wind generation scenarios were originally developed for all penetration levels by site.
While this leads to some unrealistic variability at higher penetrations due to limitations
of wind speed data, dispatch simulations were run for all of these situations. Figure 32
illustrates how the higher (albeit artificial) correlation and less geographic dispersion of
wind generation production effects integration cost.
f n e e'xCORPORATION Page 5
$30,
;.-
$25,
$20,
:r:
$15,
---
$10,
$5.
.f;
--+- eastern Montana
-11- Columbia Basin
-1..- SO/50 Mix
....;,......
Diversified Mix
700
~igure 32:Effects of geographic dispersion of wind generation facilities on integration cost. Base
case scenarios are indicated by the star symbols.
( n e e'X
100 500 600
CORPORATION Page 52
200 300 400
Total Wind Generation Capacity (MW)
Section 8
SUMMARY
The results presented in the previous discussion are the culmination of an exhaustive
and iterative pro~ess involving several hundred annual simulations of the Avista
system. Throughout the investigation all aspects of Avista operations were explored
and the data and assumptions were refined accordingly. Some new understanding of
wind integration cost drivers were developed as a result of the study. The influence of
wind generation variability and short-term uncertainty was analyzed extensively and
incorporated into the analysis. From this, new insights such as the effect of rules for
energy transactions on wind generation integration were developed and quantified. In
all, the analytical approach built on the latest developments in wind integration
analysis and then extended them significantly.
The results show that the costs for integrating significant amounts of wind
generation into the Avista power system are modest. In addition, there are
opportunities for reducing these costs. As wind generation continues to grow in the
Pacific Northwest, mechanisms for managing the additional variability and uncertainty
will be explored and implemented. As reported here , the integration costs reflect
current-day assumptions and rules for Avista system operation.
HIGHER WIND PENETRATION EQUALS HIGHER INTEGRATION COST
The Avista study confirms what other studies before it have theorized or shown through
analysis. Higher wind penetration levels , all other things being equal, increase wind
integration costs. To provide a full understanding of wind integration costs , this study
ran the LP Model through varying levels of wind penetration, from five percent up to
approximately thirty percent. This wide range covers where many systems are today,
and pushes the envelope well beyond the 20% level mentioned by many as an upper
bound for wind penetration.
INTEGRATION COSTS ARE CORRELATED WITH MARKET PRICES
Capacity opportunity costs are a significant component of wind integration. As prices
rise, all things equal, one might expect integration costs to rise as well. Wind resource
value, therefore , does not rise equally with the market price, as integration costs
consume some of the additional value. Avista used the LP Model to look at two price
sensitivities - market prices equal to half of forecasted levels, and twice forecasted
levels - and found that market prices and wind integration costs are correlated.
SHORTER-TERM MARKETS CAN REDUCE COST OF VARIABILITY
In this study, the increased short-term uncertainty due to wind generation forecast
errors increased the amount of reserve capacity required to operate the system. Much
of this is driven by rules that govern short-term exchanges of energy in the Pacific
t n e e'XCORPORATION Page
Northwest. Because the "window" for hourly trading closes well in advance of the hour
probable errors in wind generation forecasts become significant.
While improvements in wind generation forecasting can assist, reduction of the lead
time for energy transactions would also have an influence. In regions with well-
functioning short-term energy markets (some cleared at intervals as short as 5
minutes), variability in demand due to both wind generation and load variability is
spread out over a much larger footprint. When the aggregation effects on variability
over this larger geographical area are considered, the net effects on system operation
can be substantially reduced.
RISING FORECAST ERROR INCREASES INTEGRATION COST
Forecast error affects the overall level of reserve capacity necessary to integrate wind
resources. As forecast error rises, so do integration costs. Many participants to the
wind integration debate disagree on how accurate wind forecasts , and hence forecast
error, are. This study strives to identify an appropriate level of reserves to account for
forecast error; the debate will continue. To this end, Avista ran its LP Model under
various levels of forecast error, from zero percent, or perfect foresight, to thirty percent.
GEOGRAPHIC DIVERSITY HAS DIRECT INFLUENCE ON INTEGRATION COSTS
Additional generation capacity must be reserved to manage increased control area
variability and uncertainty. This capacity is a major component of integration cost.
Wind plants concentrated in a small region will exhibit a much higher degree of
correlation in their output than plants separated by larger geographic distances.
OPERATIONAL COORDINATION BETWEEN THE CONTROL CENTER AND WIND
GENERATORS CAN REDUCE INTEGRATION COSTS
There can be times where the incremental cost for managing wind generation rise
dramatically. In these times, the most economic solution may be to "feather" wind
energy via production curtailments.
t n e e'XCORPORATION Page 54
Section 9 REFERENCES
(IJ
l2)
l3)
(4J
(5)
(6J
(7)
(8J
Utility Wind Interest Group (UWIG): "Characterizing the Impacts of Significant
Wind Generation Facilities on Bulk Power System Operations Planning" May,
2003 www.uwig.org
Hirst , E. and Kirby, B. "Separating and Measuring the Regulation and Load
Following Ancillary Services" November, 1998 (available at www.EHirst.com)
Hirst, E. and Kirby, B. "What is the Correct Time-Averaging Period for the
Regulation Ancillary Service?" April, 2000 (available at www.EHirst.com
Piwko, R., et.al. "The Effects of Integrating Wind Power on Transmission System
Planning, Reliability, and Operations - Report on Phase 1: Preliminary Overall
Reliability Assessment" for the New York State Energy Research and
Development Authority (NYSERDA), published February, 2004 (available at
www.nvserda.org/ energyresources /wind. htm1)
NRELjCP-500-26722: "Short-term Power fluctuation of Wind Turbines:
Analyzing data from the German 250 MW Measurement Program from the
Ancillary Services Viewpoint"
Parsons, B., et. al. "Grid Impacts of Wind Power; A Summary of Recent Studies
in the United States" presented at the 2003 European Wind Energy Conference
Madrid , Spain, June 2003.
Milligan, M.R. "A Sliding Window Technique for Calculating System LOLP
Contributions of Wind Power Plants" prese~ted at the 2001 AWEA Windpower
Conference, Washington, DC, June 4-2001. NRELjCP-500-30363
Milligan, M., et. al. "An Enumerative Technique for Modeling Wind Power
Variations in Production Costing" presented at the International Conference on
Probabilistic Methods Applied to Power Systems , Vancouver, BC, Canada
September 21-, 1997. NRELjCP-440-22868
(9J Milligan, M., et. al. "An Enumerated Probabilistic Simulation Technique and
Case Study: Integrating Wind Power into Utility Production Cost Models
presented at the IEEE Power Engineering Society Summer Meeting, Denver, CO
July 29 -August 1 1996. NRELjTP-440-21530
Milligan, M.
, "
Measuring Wind Plant Capacity Value" NREL White Paper
Milligan, M. "Windpower and System Operation in the Hourly Time Domain
presented at the 2003 AWEA Windpower Conference , May 18-, 2003 , Austin
TX. NRELjCP-500-33955
(10)
(11J
(12J Hirst, Eric
, "
Interaction of Wind Farms with Bulk Power Operations and
Markets" prepared for the Project for Sustainable FERC Energy Policy,
September 2001
f n e e'XCORPORATION Page 55
(13J
l14J
lIS)
(16)
Milligan, M.R. "A Chronological Reliability Model to Assess Operating Reserve
Allocation to Wind Power Plants" presented at the 2001 European Wind Energy
Conference, July 2-, 2001 , Copenhagen, Denmark. NREL/ CP-SOO-30490
Milligan, M.R. "A Chronological Reliability Model Incorporating Wind Forecasts
to Assess Wind Plant Reserve Allocation" presented at 2002 AWEA Wind power
Conference, June 3-, 2002, Portland, OR. NREL/CP-SOO-32210
Karady, George G., et. aI.
, "
Economic Impact Analysis of Load Forecasting
IEEE Transactions on Power Systems , Volume 12 , No., August, 1997. pp.
1388 - 1392.
L.L. Garver, Effective Load Carrying Capability of Generating Units IEEE
Transactions on Power Apparatus and Systems VOL PAS-, No 8, pp 910-919
August, 1966
~ n e e'XCORPORATION Page 56
ApPENDIX A
WIND GENERATION
CONTRIBUTION TO PLANNING MARGIN
This section explains how the majority of wind integration costs is created by the
consumption of reserve capacity products , namely regulation, load following, and
forecast error. Each of these products is met by other resources with "quick-start"
capabilities. While wind is unable to self-provide these quick-start reserve capacity
products , it does appear capable of meeting another key capacity product-on-peak
generation capacity.
BACKGROUND
On-peak generation capacity is the contribution of a given resource to meeting system
requirements during the highest load hours of the year. Most traditional resources
provide contributions near their nameplate capacities. For example, a coal-fired plant
would be expected to generate approximately 90% of its nameplate capacity during on-
peak periods. Hydroelectric plants can provide a nearly 100% contribution during peak
load times. Wind generators, due to their limited and unpredictable fuel supply, have a
much lower on-peak capacity contributions.
Resource planners tabulate the on-peak capacity of their portfolios and compare them
to expected peak loads. Peak load is subtracted from the total of on-peak resource
capacity to determine a utility s position. On-peak resource capability must equal or
exceed expected on-peak load in a reliable system. In fact, given reliability
considerations, on-peak resource capability is expected to exceed on-peak load by an
additional planning margin. In California, regulated utilities are obligated to a planning
margin level of between 15% and 17%. Recent work by the Northwest Power and
Conservation Council (NPCC) identifies both winter and summer planning margin
targets of 25% and 17%, respectively for the Northwest.
Resource planners account for the impact of wind generation in their respective
capacity plans. Various methods exist to estimate resource contributions to system
peak periods. Some are more data- and time-intensive than others. Avista for this
report chose the Energy Load Carrying Capability (ELCC) method to evaluate wind
generation on-peak capacity contribution. The ELCC method is fairly straight-forward.
Generation at a given plant is tracked during historical peak hours as a percentage of
nameplate capacity. The resuits of this analysis are then used to estimate the on-peak
capacity contribution of the resource.
DATA AND ANALYTICAL METHOD
Avista analyzed wind data from the BPA Long Term Wind Database over a 16-year
period ending in 2004. This period of record was selected because Avista has ready
t n e e'XCORPORATION Page 57
access to its area loads on an hourly basis beginning January 1 , 1989. Five wind
locations across the Northwest were considered both individually and in combination to
understand the benefits of geographical dispersion to on-peak capacity contribution.
Hourly wind generation values based on the OSU database at each wind location, and
for all of the locations combined, were matched up with historical hourly Avista loads.
For each year studied , the top 10 and 100 hours in both the summer (July through
September) and winter (November through March) periods were evaluated. Blank data
points , where no data existed for the wind location, were ignored.
ReSULTS
The results of the top 10 and top 100 load hours had similar results, so the lOa-hour
data are presented in this report. Additionally, given that some wind data was missing,
using only the top 10 load hours in each year resulted in many fewer data points to
examine. The ELCC analysis found large differences between wind locations, and also
between the winter and summer. For example, Browning Depot, MT, provided an
average ELCC contribution in the summertime of approximately 14%; in the summer
the value was slightly higher than 41 %. Goodnoe Hills , in Klickitat County, WA, had a
higher summer ELCC, at 32%, but a lower winter rating of 14%. The following table
details average results of the Avista ELCC work.
70%
D Summer U Winter
.?:- 60%
t:5 50%
140%
...
0 30%
~ 20%
CI..
u 10%
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -------------- - ~- -
Browning
Depot
Cape
Blanco
Goodnoe
Hills
Kennewick Sevenmile
Hill
Site
Average
Figure 33: Average Summer and Winter ELCC Contributions
Avista believes that using average ELCC results for capacity planning is inappropriate
for 5 reasons: 1) a relatively small base of wind resources presently located in the
Northwest; 2) Northwest generation is not as geographically diverse as shown in the
Avista analysis; 3) the lack of Northwest utility operating experience; 4) the low on-peak
f n e N e'XCORPORATION Page 58
capacity contribution exhibited by the Northwest wind fleet over the past 2 years; and 5)
the reality that Avista s wind fleet will not be diverse for at least a period of 10 years.
Average results from the ELCC prove interesting; however, the variation in results
across the 16 evaluated years is significant. Figure 31 details results of the same wind
resources, but provides the minimum and maximum annual ELCC values for each
during the winter months , the traditional peaking period of both Avista and the
Northwest.
70%
.. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. ..
Maximu
.. .. .. ..
Average
....~.... ..
Minim
~~~ .. .. .. .. .. .. .. .. .. .. "' ..,.. :':..;.
P:'
060%
III
~ 50%
III
g. 40%
IIIc::
...
030%
c::
g, 20%
III
~ 10%
iJJ
Browning
Depot
Cape Blanco Goodnoe
Hills
Kennewick Sevenmile
Hill
Site
Average
Figure 34: Winter ELCC Contribution Distributions 1989-2004
The 5-site 16-year average of 28% is bounded by a minimum annual value of 19% and
a maximum annual value of 36%. Over the period the regional look at wind would
explain that the 100-hour average value in the worst year was nearly a third less than
the average. Additionally, the ELCC look is a 100-hour average contribution.
Individual hourly or even daily conkibutioilS will necessarily be less.
ApPLICATION TO AVISTA RESOURCE PLANNING
Avista probably will never have the full benefit of the 5-site diversity. It is likely to take
Avista many years to procure enough wind generation to make geographic diversity real.
Additionally, transmission constraints likely will preclude the utility from acquiring
wind sited in high on-peak capacity factor Montana for many years. There also appears
to be significant opposition to wind generation located on U.S. coastlines, where the
high on-peak capacity factor Cape Blanco resource resides.
It is most likely that Avista will acquire wind in the Columbia Basin, where the majority
of Northwest wind presently is being generated and sited. Three of the 5 sites evaluated
by Avista are located in the Columbia Basin: Goodnoe Hills, Kennewick, and Sevenmile
Hill. The average capacity factor of these resources is 17 percent over 16 years, with
the simple average minimum generation level equaling 10%. Individually, the on-peak
contribution falls to a low of 3% for Goodnoe Hills. ELCC during 6 of 16 years at
Goodnoe Hills is below 10%; 2 years are below 5%. ELCC at Sevenmile Hill is below
f n e e'XCORPORATION Page 59
10% in 5 of 16 years. Kennewick has larger average and minimum ELCC levels
bringing up the 3-site average.
The variability of ELCC statistics over time and location concerns Avista, especially in
light of the fact that it will be a number of years before Avista is taking generation from
more than one wind site. On one hand it is unreasonable to ignore the on-peak
contribution of wind generation entirely. On the other it is equally unreasonable to rely
on a diversified mix of sites averaged over 16 years when defining an on-peak capacity
contribution. Avista believes that future resource acquisitions should evaluate wind
generation on-peak capacity contribution on a case-by-case basis, using the lowest
annual average ELCC value. Resource capacity planning, as stated before , is intended
to protect against adverse conditions. Average values are overly optimistic for adverse
planning. The average annual ELCC still exposes the utility to some risk of lower-than-
planned-for wind contribution, but hedges this risk by picking the low-end range of on-
peak capacity contribution. For Integrated Resource Planning, where future wind
acquisitions are theoretical and not tied to any specific basin, Avista will assume an on-
peak wind capacity value of zero. This decision is based on the large number of low on-
peak contributions found in the 3 Columbia Basin locations , as well as recent
experience over a few high load conditions where regional wind generation was very low
or non-existent, and Avista s share of The Stateline Wind Generation Facility produced
no power.
t n e e"xCORPORATION Page 60
ApPENDIX B
ADDITIONAL CHARTS AND TABLES
Market Prices -- High Price Case
.:.
r."
.,,.'.'
~veraQe~ater"
../, ..'.',,;
, LpYf'Nater:,
.." "."..;".."
HiQJtwater
...,..,
MQnt~,, Peale:;'o Off-Pecle', Flat' .;, , Peak" Off-Peak: ,"".' Flat ,., . Peak'
, '
' Off~Peak,d ,
Flat, .
Jan-07 $163..43 128.148,153.135.146.79..41 84.81.78
Feb-07 $168.143,157.167,140,155.118.108,114.
Mar-07 $155.149.152,153.142.148.122.106,116.
Apr-07 $115.89.103.121..59 92..45 109.62.37.51..84
May-07 $131..39 92.114.110.66,91.
Jun-07 $47.28.39.7710 49.65.77
Jul-07 $30.16.24.95.69,84,75.45.62.
Aug-07 $123.07 101..01 113.135.99.120..45 128.104.118..44
Sep-07 $155,111..00 136.139.121..11 131..64 136.111.125.72
Oct-07 $114..42 115.114.136.113.126.127.106,118.
Nov-07 $131.115.124,151..88 128.142.152,125.72 141..22
Dee-07 $135.112.125.185.171..19 179,149.141..96 146,
Average $ 122.100.113.135.110.125.97.82.90.
Market Prices .. Low Price Case
A;verage;wate"
..."
LowWater ,
" ,.' ""' ..;,
HighWa~er
Month ,/ ,P'eaki Off-Peak'
, ,
Frat ", Peak , Off~Peak', Flat" '
. "
Peak, .' Off-Peak," Flat
Jan-07 $40,32,37.38..49 33.36.19.85 21..23 20..44
Feb-07 $42..06 35.39.41..92 35.38.29.27.28.
Mar-07 $38.37,38.38..40 35,37.3014 26,29,
Apr-07 $28.22.25.30..40 23,27.15,9..48 12,
May-07 $32.23.28.27.1610 22.2..48 1..55
Jun-07 $11.19..42 12..47 16..44 1..64 1..39 1..53
Jul-07 $23.17..48 21..09 1819 11.15,
Aug-07 $30,25.28..40 33,24,30.32.26.29.
Sep-07 $3817 27.75 34,34.30.32,34.27.31..43
Oet-07 $28.28,28.34.28.31,31.26.29.
Nov-32.28.31.37.32,35.38.31..43 35.
Dee-33,28.31..48 46..49 42.44.37..40 35..49 36.
Average 30.25.28.33.27.31.29 24.20.22.
t n e N e'XCORPORATION Page 6 1
Market Prices -- Low Spread Case
Jan-07 77,69.74.78.66.49 73.38.44,40,
Feb-07 81.45 75.78,82.71.93 77.57.57.57.
Mar-07 77,75.76.74,74,74.44 60.54.58..Q3
Apr-07 54.73 48,51.97 55.54.70 54.29.20.25,
May-07 61,51.70 57.50,39.45,
Jun-07 21.17.19,37.26.32.
Jul-07 13,10.12.43,41,42.35.45 24.31..16
Aug-07 59.53,56.65.53.60.60,57,59.
Sep-07 72..82 61.68.67.64.65.66.58.62.
Oct-07 57.57.44 57.65,59.63.61.42 56.59,
Nov-07 63,59.62,73.68.40 71.01 73.66.70.
Dee-07 65.45 59.62.91.97 86.89..82 73,72..73 73,
Average 58.53.56.65.47 58.62.47.43.45.45
Market Prices .. High Spread Case
:i.::i Averag~wafe"':"ii.:(ii:' 'loW'Wafer
, ."" ::. ,' "
':i.HighWafer' . ,Monf~,
. ", :
l'eak Off"Peak 'Flaf/
, ,
Peak Off.'Peak Flat
. .
Peak Off-Peak . Flat
Jan-07$86.57.74.79.64.73,42.37,40.
Feb-07 85,69.73 78.77 88.64.77.61.18 51.57.
Mar-07 85.47 62.76.49 89.45,74.44 67.79 43.58..Q3
Apr-07 64.33.51.97 70,30.54,36.11.93 25.
May-07 77.28.57,58.75 29.45.79
Jun-07 28,19.45.49 15.32.
Jul-07 16.12.51.48 30.42.41.17,31.16
Aug-07 69.39.56.73.43.60.68.47.59.
Sep-07 77.55.68.74.79 53.65..82 73.49.62.
Oet-07 60.52.57.73.49.63,67.47.59,
Nov-07 70.42 50.62,77.62.45 71.79.59.70.
Dee-07 73.49.62.99,76.40 89.80.61.73,
Average 66.42.56.73.47.62.52.36.45.45
f n e e'XCORPORATION Page 62
Market Prices by Water
Year
Jan-07 81.64..0 1 74.76.67.75 73.39,42..47 40,
Feb-07 84.71.78..77 83,70.77.59.54.57.
Mar-07 77.74.76..49 76.71.74..44 61..48 53..44 58..Q3
Apr-07 57,44,51.97 60.46,54.31.13 18.25.
May-07 65.70 46.57.55.33..41 45.
Jun-07 23.14.19,38.24.32.2.78 3 ..07
Jul-07 15.12.47.34.42.37,22.31.16
Aug-07 61.50,56,67,49.60.64..48 52.59.
Sep-07 77.55.68.69..77 60.65.68.55.62.
Oct-07 57.57.57.68.56.63.63.77 53.59.
Nov-07 65,57.62.75.64..43 71.01 76..42 62.70.
Dee-07 67.56.62.92.85.89.74.70.73.
Average 61.35 50.56.67.55.62.48.41.45.
t n eCORPORATION Page
ApPENDIX C
NEXT STEPS
In early research performed to prepare for this study, it was not always easy to
understand the many differences , strengths and weaknesses of each work. This made
comparing the methods and their merits difficult. For the benefit of future studies
Avista highlights below areas of its study that warrant additional consideration in
future work. As with any study performed in a fairly new field, it is almost impossible
to be certain that an outcome is all-inclusive. This said, Avista is confident that the
results presented in this report are substantially correct in the total and cannot at this
time be certain that re-visiting these issues in total will lead to either higher or lower
integration costs.
This study applies the latest methods of wind integration analysis. In reviewing the
final work product, the authors would like to acknowledge that further work should be
performed in the following areas:
CALCULATING RESERVES
Wind integration costs stem primarily from incremental system reserves necessary to
balance instantaneous output with power schedules. This study identified incremental
reserves for regulation, load following, and forecast error. The method for calculating
regulation was based on a "5-sigma" approach applied to one-minute data.
Forecast error was calculated for wind by using an average of wind generation from 60
to 120 minutes prior to the delivery hour. This method was applied to reflect the real-
time scheduling window. Forecast error was reduced by 25% to reflect actual statistics
the Company has witnessed from "state-of-the-are" wind forecasting techniques. Avista
does not record its hour-ahead load forecasts so that forecast error of combined wind
and load may be calculated. Wind forecast error was reduced by a further 15 MW (1. 5%
of average hourly load) to reflect an estimated load forecast error level.
Forecast error and load following represent machine flexibility that must be reserved
and do not appear correlated. This lack of correlation allows a lower combined level of
reserves to be held. Load following in this report was determined as the combined load
following and forecast errors less the forecast error calculated above. This method
necessarily overstated forecast error and understated load following reserves. However
together the two values represent the appropriate level of total intra-hour reserves.
Load following and forecast error reserves were calculated in a method that represents
the best thinking today. For this study the calculations were broken into ten deciles
based on the capacity factor of the wind. This approach was found to reduce overall
forecast error and load following reserve levels modestly.
f n e e'XCORPORATION Page
Further discussion over the best means to approximate how operators would schedule
load following and forecast error reserves is warranted. Given the high degree of
correlation between operating reserve levels and integration costs, future studies should
look for new ways to reduce the necessity of such reserves.
WIND GENERATION DATASET
Wind generation datasets for the Northwest are very limited. Absent a robust dataset
this study relied on a set of anemometer data collected by the Bonneville Power
Administration. Six individual anemometers located across the Northwest were used as
the basis for the study s wind data. There appear to be methods to approximate
diversification of a wind farm in the regulation time frame using a single anemometer;
however, the "science" of extrapolating a single anemometer to a larger wind farm over
an hourly or multi-hour period is less known, especially in cases where large wind
quantities were analyzed.
Avista is reasonably confident that its datasets provide a good representation of wind
generation in its Base Case scenarios. However, some of the large single-basin
penetration levels used in the study likely overstate wind variability due to reliance on a
single anemometer to represent the wind regime. Avista looks forward to the wind data
expected to become available in 2007 or 2008 in response to an action item in the
recent Northwest Wind Integration Action Plan. This data will be used to enhance the
Avista work once it becomes available.
THERMAL GENERATION MODELING
The Avista study did not model all of the costs associated with its thermal resources.
Starting Coyote Springs 2, a combined-cycle combustion turbine, costs the company on
the order of $20 000. This start-up cost will limit the hourly dispatch of the resource
and likely will increase wind integration costs. Our other thermal plants also witness
similar costs that are not modeled in the current Avista analysis.
Avista owns 15% shares in two coal-fired power plants located in Montana. These
resources were not modeled to provide any reserve products given their position in the
resource stack and modest reserve capabilities.
These modeling simplifications likely did not impact the results of the Avista study in
any significant way. They enabled the Avista LP Model to solve more quickly and
thereby enabled significantly more scenarios to be evaluated. Future integration
studies based on the LP Model will consider enhancing thermal plant logic to better
represent costs.
TRANSMISSION
The LP Model contains detailed transmission logic. All energy (purchases and sales)
and reserve (transfers from remote Avista resources to its load center) transfers
occurring at or below Avista contract rights do not pay any transmission tariff, as these
~ n eCORPORATION Page 65
costs are "sunk." Only system losses are charged for energy moving across the grid.
Any hourly transmission quantities in excess of existing contract rights pay both losses
and the hourly cost of firm transmission.
New wind resources are assumed to have come with firm transmission paths to Avista
system. The assumption lowers wind integration costs stemming from wind variability
since no new transmission bottlenecks are created by the inclusion of wind. It might
not be appropriate to assume a one-for-one purchase of transmission. This assumption
will be re-evaluated in future analyses.
f n e e'XCORPORATION Page 66