1. Hybrid Forecast for Resource Adequacy Analysis with recommendations Massoud Jourabchi April 21 2016

2. Load forecasting Models used What is Long-term model (LTM) What is Short-term model (STM) What is the hybrid model What are the recommendations for creating load forecast for future RA

3. LTM Designed to provide a 20 year forward look for use in Conservation and DR Assessment Regional Portfolio Model Time resolution for the forecasts is monthly. Assumes normal weather in the future. Produces three different Load forecasts (Price effect, Frozen-efficiency and Sales forecast) Quarterly Frozen-efficiency forecast is provided to RPM for resource selection. Sales forecast*, is Frozen-efficiency loads net of Energy Efficiency Annual sector and enduse level EE incorporated into the Frozen-efficiency model, so that monthly shape of EE is part of Sales forecast. *- although labeled sales forecast, this is a load forecast (at the generator busbar)

4. STM Designed to produce hourly forecast of regional loads, with a 3-5 years forecasting horizon. Incorporates impact of temperature on load on an daily and hourly basis. It does not make a forecast of future weather regimes but instead uses past daily regional temperatures in creating future hourly loads. It creates weather normalized daily load forecast, and using hourly allocation factors it generates hourly WN load for the target year. It then subjects to hourly WN loads to hourly multipliers that capture hourly temperature sensitive loads to the hourly WN loads.

5. Overview of Analytical Steps in STM Starting with daily temperatures we estimate the normal or average temperature for the day and the deviations from these temperatures for each day since 1928. Temperature deviations along with daily regional load and a number of other explanatory variables are used to estimate the structural relationship between daily load and daily temperature. The structural relationship is then parsed into two parts. 1)Weather normalized daily load 2)Temperature variables that capture the relationship between load and daily temperature deviations from normal A Daily load forecast under the 86 past historic daily temperature is created. To create an hourly load forecast, an hourly model is created looking at the normal hourly temperature, the deviation in hourly normal temperature. Then the deviation from normal hourly temperature along with hourly loads and other explanatory variables are used to establish to create a matrix of hourly allocation factors. The daily load forecast and the hourly allocation factors are combined to create an hourly load forecast under 86 different past weather regimes.

6. Differences between LTM and STM LTMSTM Intended Applications20 year, Conservation tracking enduse efficiency Policy and forecast model 3-5 year forward look, Resource Adequacy Methodology differencesFrozen Efficiency, knows about future codes and standards, Econometric modeling, imbedded Energy Efficiency, no knowledge of future policies codes/standards. Impact of weatherNormal weather for futureExplicit account of hourly past temperature conditions FocusMonthly EnergyHourly Energy and Peak Data updateEvery 5 years, by sector, enduse, technology, by state Annual, region-wide

7. Hybrid Approach WN Monthly LTM F.E.* Load net of EE (LTM Sales) Daily and Hourly WN load allocation factors applied to WN monthly load Creates hourly WN loads for the 2021 Add in 86 different hourly profiles to the WN base Creates hourly Temperature Sensitive Loads 2021 load forecast net of EE Hourly load under the 86 different past different regimes This approach replaces the WN loads from STM with the WN loads from LTM load forecast. WN loads from the LTM can only be fully updated every 5 years, as part of Power Plan. For RA application we can update the temperature, economic drivers, natural gas prices of the LTM every year. EE amount will not be updated for RA analysis.

8. Recommended approach for the future Change sampling period from 1929-2005 to 1993-2014 or the latest year available. Use hourly Short-term model instead of daily model and hourly allocation factors. Subject the WN load forecast to capture a wider range of loads, similar process used in RPM.