20150818AVU to Staff 79.doc

AVISTA CORPORATION
RESPONSE TO REQUEST FOR INFORMATION

JURISDICTION: IDAHO DATE PREPARED: 08/12/2015
CASE NO.: AVU-E-15-05/AVU-G-15-01 WITNESS: Tara Knox
REQUESTER: IPUC RESPONDER: Tara Knox/Lorin Molander (DNV GL)
TYPE: Production Request DEPARTMENT: State & Federal Regulation
REQUEST NO.: Staff - 079 TELEPHONE: [phone]

REQUEST:

Pages 17 through 20 of the Electric Load Research Study presented in Ms. Knox' workpapers describe a method for editing the data. The editing method included processes for eliminating suspected outliers and for recreating missing data. Please answer the following:
a. How were the spike elimination and data recreation methods validated?
b. How sensitive are the estimates of load factors, coincident peaks, and non-coincident peaks to DNV-GL's data editing methods?
c. Would use of unedited data in DNV-GL's model have resulted in a significant (>1%) difference in estimated load factors, coincident peaks, and non-coincident peaks?

RESPONSE:

As described on page 19 of Ms. Knox’ workpapers, “The third editing step was to reexamine each individual site using Visualize-ITTM. This examination compared the original and filled data for the site.” After the spike elimination and data recreation steps, data with filled intervals was imported into DNV GL’s Visualize-ITTM software to visually validate that the intervals were filled with acceptable levels of demand. Acceptable levels of demand are defined as estimated demands following similar characteristics as actual demands (e.g., weekend vs. weekday patterns, hourly time of day use patterns, etc.). Unacceptable estimated demands are demand patterns that do not behave in the same manner as a site’s actual demand pattern in other similar time periods, or estimated demand levels that are much lower or greater than what was actually achieved by the site in other time periods. This step is a manual process requiring judgment, and it was performed for Avista by an analyst with 20 years of experience working with energy load profile data.

It is unknown exactly how sensitive the estimates of load factors, coincident peaks, and non-coincident peaks are to DNV GL’s data editing methods. Following industry best practice, DNV GL’s methodology does not create artificial peaks (at a site) with edited data.

As shown in the Table 12 on page 20 of Ms. Knox’ workpapers (a copy is provided here for reference), a total of 5,088 missing hourly intervals were estimated and filled for Avista’s Idaho jurisdiction. These intervals are from two residential sites. It is unlikely that there would be a material difference between the load factors, coincident peaks, and non-coincident peaks using the edited interval dataset versus the unedited interval dataset in the model. In this case, we believe the load factors and peak statistics calculated would not be sensitive to DNV GL’s editing methods.

As discussed in (b), it would be highly unlikely there would be any significant difference in the estimated load factors, coincident peaks, and non-coincident peaks if the unedited data had been used in the model for Idaho.

Page 2 of 2