HomeMy WebLinkAbout20240814IPC to Staff 34 - Attachment 1 - Weather Normalization Methodology.pdf Response to Staff Request No.34-Attachment 1
IDAHO POWER
SUPPORTING DOCUMENTATION
WEATHER NORMALIZATION
L Weather Normalization Methodology Overview
To determine the degree to which actual electricity sales were higher or lower than normal as a result
of actual weather, it is necessary to first quantify the relationship between weather and sales. This
quantification is achieved in the IPCo model through the use of multiple regression) analysis in
which energy use is statistically estimated as a function of weather and nonweather variables. This
relationship between electricity use and its primary determinants is measured in regression equations
- IPCo's total system residential and commercial sales; IPCo's Oregon residential and commercial
sales; five which describe irrigation sales for each IPCo; IPCo's Oregon irrigation sales.
To explain electricity use,the regression equations utilize weather concepts such as heating, cooling
and growing degree days and precipitation, in some cases economic information such as real
electricity prices or total area employment, is used. As indicated in the detailed methodology
description, the summary statistics demonstrate that the regressions are accurate in explaining
monthly variations in sales. This is particularly true of the weather variables. Combining data from
many months in each regression and estimating over a long time period allows sufficient variation
within the data to incorporate economic variables, if needed, and more than one weather concept.
The residential and commercial equations quantify use per customer as a function of weather and
nonweather variables while the irrigation equations explain electricity use by operating center.
Once the regression equations have been specified and estimated,it is the coefficients of the weather
variables that are of primary importance to the weather adjustment process. These coefficients
measure the response of sales to changes in those weather variables. For example, the coefficients
of the heating degree day variables in the residential total system equation represents the number of
KWh/customer that one additional heating degree day would cause. By multiplying this coefficient
by the difference between the normal number of heating degree days for a particular month and the
number that actually occurred, the difference between actual and normal KWh/customer is
'A functional relationship between two or more correlated variables that is often empirically determined from
data and is used especially to calculate values of one variable when given values of the others.
determined. It is important to recognize that the non-weather variables are used only to estimate the
regression equations and that only the coefficients of the weather variables are used in the actual
adjustment. It is also important to note that the primary purpose of the models is to adjust sales that
have already occurred, rather than to estimate future sales. Although the equations could be utilized
as forecasting tools,the focus of these models at the present time is on their ability to adjust historical
sales for abnormal weather, not their ability to forecast.
Weather Data
The weather concepts in the regressions are constructed from temperature and precipitation data
from four weather stations: Boise, Pocatello, and Twin Falls, Idaho, and Ontario, Oregon. One of
the most critical aspects of quantifying the relationship between electricity use and weather is the
correct matching of weather with sales data. Because the Company's residential, commercial, and
irrigation customers are billed in cycles, sales in a particular month represent consumption that
occurred in portions of the current month as well as the previous two months, depending upon the
particular billing cycle being analyzed. Consequently, weather that occurred during the current
month and the previous two months must be considered when examining sales in the current month.
To account for this,and to correctly match weather data with sales data,the weather variables utilized
in the residential,commercial,and irrigation regressions represent weighted sums of the daily values
of those variables over the appropriate time period. Each day's weather measure is weighted by the
number of customers (for a particular month)who experienced that weather.
The IPCo service area contains regions with different weather patterns.To incorporate these different
weather patterns in the system regressions, the weather variables were constructed using weighted
weather data from the Boise, Twin Falls, Pocatello, and Ontario weather stations. For example, the
heating degree day variable used for each month in the system residential equation is a weighted sum
of the heating degree days of Boise, Twin Falls, Pocatello, and Ontario with the weights based on
the number of customers in IPCo's five operating centers represented by those cities.By constructing
the weather variables in such a way, the model has the capability of incorporating the diversity of
weather in the Company's service area.
II Detailed Methodology Description
In general, the weather concepts used in the weather adjustment model are matched with sales as
follows:
1) For those categories representing customers billed by calendar month, calendar
month heating and cooling degree actuals and normals are collected from the National
Oceanic and Atmospheric Administration(NOAA).
2) For those categories representing billing cycle customers, (System Residential,
Oregon Residential, System Commercial, Oregon Commercial, Irrigation by Operating
Center, and Oregon Irrigation), monthly weather concepts are generated by aggregating
daily data over the appropriate days of the billing month. Weather concepts calculated in
this manner include heating degree days, cooling degree days, growing degree days, and
precipitation. Each of these weather concepts are calculated for four service area weather
stations: Boise, Pocatello, Twin Falls, and Ontario.
Weather Data Sources
The following weather data is used directly in the weather adjustment or in the construction of the
monthly billing month weather variables (the source of data is the NOAA):
Actual Weather Data
Ontario: Monthly Heating and Cooling Degree Days from 1961 Forward
Twin Falls: Monthly Heating and Cooling Degree Days from 1964 Forward
Boise: Daily Maximum Temperature, Minimum Temperature, and Precipitation- 1961 Forward
Pocatello: Daily Maximum Temperature,Minimum Temperature, and Precipitation- 1961 Forward
Twin Falls:Daily Maximum Temperature,Minimum Temperature,and Precipitation- 1964 Forward
Ontario: Daily Maximum Temperature, Minimum Temperature, and Precipitation- 1961 Forward
Normal Weather Data
Ontario: Monthly Heating and Cooling Degree Day Normals (Median)
Twin Falls: Monthly Heating and Cooling Degree Day Normals (Median)
The degree day concepts used in the weather adjustment are defined as follows:
Heating Degree Days= Maximum[0,65-((Maximum+Minimum Temperatures)/2)J
Cooling Degree Days= Maximum[0,((Maximum+Min im um Temperatures)/2)-65]
Growing Degree Days= Maximum[0,((Maximum+Minimum Temperatures)/2)-50J
Billing Cycle Daily Weights
The cyclic billing of IPCo residential, commercial, and irrigation customers has two primary
implications for the matching of weather and sales data. First, sales in a particular month represent
consumption that occurred in that month and the previous two months. Thus, a weather variable
which will be used to help explain that month's sales must reflect the appropriate period over which
consumption occurred.
Second, the impact of a particular day's weather over this period is different for different days. The
weather on a day for which consumption of all 21 cycles will be recorded for a particular sales month
is far more important to that sales month than the weather on a day for which consumption of only
1 cycle will be recorded. Consequently, when constructing monthly weather concepts from daily
data, each day's weather must be weighted according to its importance to the billing month.
Figure I demonstrates how Figure 1. Number of Cycles Recording Sales for May Sales
consumption of the 21 cycles Month
is spread over each billing
(number of cycles)
month. The diagram shows
the number of cycles 21
consuming electricity for each 18
day of the billing month based
15
on Company meter reading
dates. The month May is used 12
here for illustrative purposes. s
The first day of the first cycle
s
of this billing month is March
31 -for that day, only the first s
cycle's consumption will be o _ _
O M In n O Z M In n Q1 M IA 1� O M lf7 1� O M l2
M 0 0 0 0 0 N N N N N 0 0 0 0 0 N N N N
recorded as May sales. On aaaaao 0 0 0 0 o aaaaTTTTTTTTTTTTTT
y � a a a a a a a a a a a a a a a � � � � � � � � � � � � � �
April 1, cycle 2 begins and so
the consumption of two cycles on that day will contribute to May sales.On April 28,the consumption
of all 21 cycles will be recorded, as May sales. Those days with a higher number of cycles are more
important in constructing weather measures to explain weather-related consumption in the May sales
month than those with a smaller number of cycles.
To construct a weather measure representative of every billing month while accounting for these
differences in each day's influence, each day within the month is assigned a weight measuring the
number of cycles of that day. For example, on a day where 15 cycles are recording sales for a
particular month, the weight is 15/21. Note that all 21 cycles are consuming electricity every day.
The only thing that varies is the month in which that consumption will be recorded. Thus, if 15 of
the 21 cycles will be recorded in one month, the other 6 will be recorded in another month. The
assignment of weights to each day in this manner assumes an equal number of customers in each
cycle.
Construction of Weather Actuals
With weights assigned to each day on the basis of a number of cycles, billing month weather
variables are calculated by summing the weighted weather measure over every day in the billing
month. For example, to calculate billing month heating degree days, daily heating degree days are
weighted and summed over every day of the billing month. The same procedure is followed for
cooling degree days,growing degree days,and precipitation.The following variables are constructed
in this manner for each of four weather stations: Boise,Pocatello, Twin Falls, and Ontario and from
1961 to present(or where data permits).
Billing-Adjusted Heating Degree Days -Base 65
Billing-Adjusted Cooling Degree Days -Base 65
Billing-Adjusted Growing Degree Days -Base 50
Billing-Adjusted Precipitation
The billing-adjusted weather concepts for Ontario, Oregon are used in the Oregon-specific
regressions. The billing-adjusted growing degree day and precipitation weather concepts are used in
the operating center and Oregon-specific irrigation regressions. The system weather variables used
in the system residential and commercial regressions require one more step described below.
With these billing-adjusted weather concepts constructed for each of the four cities,it is now possible
to construct aggregate service area measures for use in the system regressions. These aggregate
service area measures are weighted averages of the billing-adjusted measures for the four cities. For
example, the residential heating degree day variable used in the system residential regression is a
weighted average of the heating degree days of Boise, Twin Falls, Pocatello, and Ontario with the
weights based on the number of residential customers in IPCo's five operating centers represented
by those cities. Similarly, the residential system cooling degree day variable is constructed as a
weighted sum of the cooling degree days of the four cities using the same weights. The system
commercial degree day variables used in the regression are calculated in the same way but the
weights used in that calculation are based upon commercial customers in the five operating centers.
Following is a list of the weather data actuals used in the residential, commercial, and irrigation
weather adjustment regressions calculated as described above:
System Residential Heating and Cooling Degree Days
System Commercial Heating and Cooling Degree Days
Irrigation Growing Degree Days and Precipitation by City for each Operating Center
Oregon Residential Heating and Cooling Degree Days
Oregon Commercial Heating and Cooling Degree Days
Oregon Irrigation Growing Degree Days and Precipitation
Construction of Weather Normals
For those customers billed on a calendar month basis, calendar month normals for heating and
cooling degree days are calculated over the years 1994 to 2023 (most recent 30 years). The median
values are calculated and used to represent a"normal" number of heating and cooling degree days.
The median figures have a 50150 chance of occurrence and are not adversely influenced by a few
extreme weather events.
For those billing cycle customers, normal weather concepts are calculated as follows:
1) For each city,normal weather concepts are calculated by first calculating the median weather
concept over the period 1994 to 2023. That is, for each city, the 30 values for January's
heating degree days are used to determine the median,as are February's values,and so forth.
This provides 12 normals for each city and for each weather concept. The normal billing-
adjusted weather concepts calculated for Ontario, together with the billing-adjusted actuals
for Ontario are used in the Oregon-specific weather adjustment models.
2)With the normal billing-adjusted weather concepts calculated for each of the four cities, it
is now possible to construct aggregate normal weather concepts representing the entire
service area. These aggregate service area normals are weighted averages of the billing-
adjusted normals for the four cities. For example,the residential system heating degree day
normal is a weighted average of the billing-adjusted heating degree day normals for Boise,
Pocatello, Twin Falls, and Ontario. The system normal weather concepts are calculated in
precisely the same manner as the system actuals. The operating center weights used for the
system residential variables are residential customers and those used for the system
commercial variables are commercial customers. Note that a normal weather concept for
January(or any other month)of one year may differ from the normal for January of another
year if the operating center weights change from year to year.
Description of Economic and Price Data
Price Terms: Each price term is an average price, calculated as the ratio of revenue to sales. In
the cases of the Oregon price terms, the revenue or sales for the appropriate
jurisdiction are used. A 12-month moving total of revenue and sales are used to
dampen seasonal fluctuations and mitigate the multicollinearity between
variables.
Electric
Space Heat
Saturation: IPCo residential survey data is used when available.
Central Air
Conditioning
Saturation: Similar to Electric Space Heat Saturation.
Total
Employment: As provided by Moody's Analytics
Description ofRe,-cessions
Total System Residential Model:
A model of electricity sales using monthly observations was estimated for the system residential
class over the period January 2008 to December 2023.
The dependent variable in the total system residential model is system residential use-per-customer.
The pattern of electricity use is described by the combination of weather and non-weather factors
discussed below.
Degree Days
The weather variables utilized by the system weather adjustment model are constructed as weighted
averages of degree day variables from four service area weather stations: Boise, Twin Falls,
Pocatello, and Ontario. That is,the heating degree day variables (Base 65)are an average of heating
degree days from the four weather stations, each weighted by the number of customers in the
operating center associated with that weather station. The same weighting mechanism was used to
construct the cooling degree day variable (Base 65).
Trends
Linear trends are used to pick up the effects of unmeasured and immeasurable factors which change
over time and influence energy consumption.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as number of daylight hours,
school vacations, holidays, etc. These variables contributed to the explanatory power of the model.
These variables could also help account for other weather-related variables, such as cloud cover and
wind speed, which the Company was not able to account for at this time but which may exhibit
definite monthly patterns. In addition a COVID binary was included in the residential model
effective Apr 2020 billing to Feb 2023 billing with diminishing impact throughout that identified
period.
Oregon Residential Model:
A model describing residential electricity sales per customer for the Oregon portion of the
Company's service area was estimated over the period January 2008 to December 2023.
The dependent variable in the Oregon residential model is Oregon residential use-per-customer. The
pattern of electricity use is described by the combination of weather and non-weather factors
discussed below.
Degree Days
The weather variables used for the Oregon specification are heating and cooling degree days (Base
65)for Ontario, Oregon.
Price
The price term represents the residential average price of electricity. The price term is constructed
from Oregon-specific revenue and sales data.
Trends
Linear trends are used to pick up the effects of unmeasured and immeasurable factors which change
over time and influence energy consumption.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as number of daylight hours,
school vacations, holidays, etc. These variables contributed to the explanatory power of the model.
These variables could also help account for other weather-related variables, such as cloud cover and
wind speed, which the Company was not able to account for at this time but which may exhibit
definite monthly patterns.
Total System Commercial Model:
The total system commercial model was estimated over the period January 2008 to December 2023
and is structurally similar to the system residential model.
The dependent variable in the total system commercial model is system commercial use-per-
customer. The pattern of electricity use is described by the combination of weather and non-weather
factors discussed below.
Degree Days
The weather variables utilized by the system weather adjustment model are constructed as weighted
averages of degree day variables from four service area weather stations: Boise, Twin Falls,
Pocatello, and Ontario. That is,the heating degree day variables (Base 65)are an average of heating
degree days from the four weather stations, each weighted by the number of customers in the
operating center associated with that weather station. The same weighting mechanism was used to
construct the cooling degree day variable(Base 65).
Trends
Linear trends are used to pick up the effects of unmeasured and immeasurable factors which change
over time and influence energy consumption.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as number of daylight hours,
school vacations, holidays, etc. These variables contributed to the explanatory power of the model.
These variables could also help account for other weather-related variables, such as cloud cover and
wind speed, which the Company was not able to account for at this time but which may exhibit
definite monthly patterns. In addition a COVID binary was included in the residential model
effective Apr 2020 billing to Dec 2021 billing was included.
Economic
Total employment for the Boise Metropolitan Statistical Area(MSA)was used as an independent or
explanatory variable in the system commercial model.
Oregon Commercial Model:
A model describing commercial electricity sales per customer for the Oregon portion of the
Company's service area was estimated over the period January 2008 to December 2023. First order
autocorrelation was detected and corrected.
The specification is structurally similar to and provides results consistent with the commercial total
system specification. The price term is constructed from Oregon specific revenue and sales data.
The dependent variable in the Oregon commercial model is Oregon commercial use-per-customer.
The pattern of electricity use is described by the combination of weather and non-weather factors
discussed below.
Degree Days
The weather variables used in the Oregon commercial specification were heating and cooling degree
days (Base 65) adjusted for billing cycles for Ontario, Oregon.
Trends
Linear trends are used to pick up the effects of unmeasured and immeasurable factors which change
over time and influence energy consumption.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as number of daylight hours,
school vacations, holidays, etc. These variables contributed to the explanatory power of the model.
These variables could also help account for other weather-related variables, such as cloud cover and
wind speed, which the Company was not able to account for at this time but which may exhibit
definite monthly patterns. In addition a COVID binary was included in the residential model
effective Apr 2020 billing to Aug 2020 billing was included.
Operating Center Irrigation Models:
A basic model of electricity sales using monthly observations was estimated for irrigation sales for
each IPCo operating center over the months January 2008 through December 2023.Operating center
irrigation sales are considered a function of growing degree days, and precipitation. This pattern of
electricity use is described by the combination of weather factors discussed below.
Degree Days and Precipitation
The weather variables utilized by the system irrigation weather adjustment model are constructed as
weighted averages of degree day variables from four service area weather stations:Boise,Twin Falls,
Pocatello, and Ontario. That is,the growing degree day variable(cooling degree days Base 50)is an
average of growing degree days from the four weather stations, each weighted by the share of total
system irrigation pumping horsepower connected in the division associated with that weather station.
The precipitation variable was constructed and weighted in the same manner.A second precipitation
variable was used to measure the impact of pre-growing season precipitation on the early months of
the growing season (May and June). This variable is zero for all months except May and June of
each year where it is the total of the weighted precipitation for the two previous months.For example,
the pre-growing season precipitation for May 2023 is the total of the weighted precipitation for
March and April 2023. This variable was used because of the fact that precipitation prior to the
growing season is partially retained by the soil. This may minimize the need for irrigation water
applications in the early months of the growing season. The opposite would also be true. Small
amounts of pre-growing season precipitation cause decreased amounts of retained moisture in the
soil which may increase early growing season irrigation applications.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as the stage of the growing
cycle that the crops are in at various times during the year, harvest patterns, etc. The linear trend is
used to pick up the effects of unmeasured and immeasurable factors which change over time and
influence energy consumption.
Oregon Irrigation Model:
A model describing irrigation electricity sales for the Oregon portion of the Company's service area
was estimated over the period January 2008 through December 2023. The specification is
structurally similar to and provides results consistent with the Operating Center models. The
dependent variable in the Oregon irrigation model is Oregon irrigation sales. The pattern of
electricity use is described by the combination of weather factors discussed below.
Degree Days and Precipitation
The weather variables used in the Oregon irrigation specification were growing degree days(cooling
degree days Base 50) and precipitation adjusted for billing cycles for Ontario, Oregon. The degree
days and precipitation variables were constructed using Ontario, Oregon historical weather data. A
second precipitation variable was used to measure the impact of pre-growing season precipitation on
the early months of the growing season (May and June) and was constructed as described in the
Operating Center Irrigation Models above.
Monthly Binaries
Monthly binary variables were included in the regression to account for factors which vary from
month to month but which are relatively constant over the years, such as the stage of the growing
cycle that the crops are in at various times during the year, harvest patterns, etc. The linear trend is
used to pick up the effects of unmeasured and immeasurable factors which change over time and
influence energy consumption.
The Use of Regression Results to Adjust for Abnormal Weather
The method by which the preceding regression results are used to weather normalize monthly sales
is the same method used by the Electric Power Research Institute as described in Weather
Normalization of Electrici . Sales.' That description is paraphrased below:
Any model that can be used for weather normalizing monthly sales can be written in the general
form:
S. - f(W. , X.)
where Sm is electricity sales per customer for month m;
w,,, is a vector of weather measures related to electricity sales
in month m;
x,T, is a vector of non-weather variables related to electricity
sales in month m; and
'Electric Power Research Institute, Weather Normalization of Electrici, Sales, (June 1983), EA-3143,
Research Project 1922-1.
f is a function of observed explanatory variables wm and x..
Given a model of this kind,weather normalization of electricity sales proceeds as follows. Predicted
A
sales per customer for month in under actual weather conditions will is
Sm- f(W'M,.Xm)
N
Predicted sales per customer for month in under normal weather conditions Wm is
Sm .l (wm , .xm) ,
N
where Wm is a measure of normal weather in month m. Consequently, the predicted adjustment in
sales that is required to reflect non-normal weather conditions is
Am m Sm) x m ,
where Cm is the number of customers billed in month m.This adjustment,Am,is applied to the actual
sales for the month of m to obtain "weather normalized sales." If predicted sales under normal
weather exceeds predicted sales under actual weather, then the adjustment is positive and weather
normalized sales are greater than actual sales. If predicted sales under normal weather is less than
predicted sales under actual weather, then the adjustment is negative and weather normalized sales
are less than actual sales.
This procedure for normalizing sales has two implications regarding the construction and evaluation
of weather normalization models. The first implication is that the primary purpose of the models is
for adjusting previous sales,rather than forecasting future sales. That is,the models are used to adjust
sales that have already occurred, with the adjustment being the portion of those sales that were due
to abnormal weather. While the models could also perhaps be used to predict sales that would occur
in the future, either under normal or predicted actual weather,that is not their primary function. This
means that in constructing and evaluating the models, their ability to accurately adjust past sales for
abnormal weather is the focus,not their ability to forecast accurately into the future.
The second implication, which follows the first, is that variables that might affect sales, but do not
affect the weather adjustment of sales,are not essential to the model.For example,consider a linear
model with two variables:
Sm a Wm + 0 xm
where w,,, is a weather variable, and
xm is a non-weather variable.
With this model, estimated sales under actual weather is
Sm a Wm + e.xm
and under normal weather is
Sm a Wm + e.xm
Consequently,the adjustment in sales that is required to reflect non-normal weather is:
Am ` U m - Sm/ x Cm - + 8 xm)x C
(a Wm + 9 xm) - (a Wm m
(a Wm - a Wm)x Cm
a(WM - Wm)x Cm
The value of the variable xm and the value of its coefficient are irrelevant in weather normalizing
sales data; only the weather variable and its coefficient enter the adjustment procedure. This
implies that, in estimation,non-weather variables that enter linearly are important to include only if
their inclusion affects the estimated coefficients of weather variables. Non-weather variables that
enter linearly and whose inclusion does not affect the coefficients of weather variables are not
important to weather normalization.
The procedure by which the Company's residential, commercial, and irrigation weather adjustment
is allocated to specific jurisdictions and schedules is as follows:
• The Oregon and system adjustments are calculated by the methodology described above
except irrigation adjustments are not multiplied by the number of customers billed.
• The Oregon adjustments are subtracted from the system adjustments to obtain the adjustment
for Idaho.
• Each jurisdiction's adjustment is then apportioned to the following schedules:
Residential- all the adjustment is applied to Schedule 01, 06, and 84R.
• Commercial - adjustment is apportioned to Schedules 07, 08, 84C, 9S, and 845.
Irrigation - all the adjustment is applied to Schedule 24S and 841.