2013 Load Impact Evaluation Capacity Bidding Program (CBP) Steve Braithwait, Dan Hansen, and Dave Armstrong Christensen Associates Energy Consulting DRMEC Spring Workshop May 7, 2014 May

2 Statewide CBP Programs  Features of programs  Methods and validation  Ex-post load impacts (2013)  Load impacts by program, product & event  Ex-ante load impacts

May CBP Features  Capacity ($/kW) payments for nominated load  Energy ($/kWh) payments to bundled customers  Monthly load reduction nominations (MW)  Product-type options  Day-ahead (DA) or Day-of (DO) notice  Event windows (1-4; 2-6; or 4-8)  4 to 6 aggregators at each utility (a few are individual customers)

CBP Events May

5 Nominated Customer Accounts; by Utility, Year, and DA & DO Notice

Nominated Customers, by Industry Type May

7 Ex-Post Regression Model (Individual Customer Level)  Dependent variable = kWh/hour  Independent variables:  To estimate hourly event-day load impacts -- – Indicator variables for each hour of every event day  To control for weather conditions -- – CDH65_3MA and 24MA  To establish typical hourly load profile -- – Separate hourly indicator variables for Monday, Tuesday - Thursday, and Friday  To control for typical load level -- – Day-of-week indicator variables – Month-of-year indicator variables

May Ex-Post Regression Model (2)  Independent variables (continued):  Event-hour indicators for events of other DR programs in which the customer is enrolled  Summer pricing season differences – Summer defined according to tariff season definitions – Separate summer load level and hourly load profile  Day-of, morning-load adjustment to improve accuracy – Average hourly load from hour-ending 1 through 10

May Model Validation  Estimate models with event-like non-event days withheld from the sample (one at a time) and examine performance of model predictions on those days (MAPE, MPE, R Sqr)  Model variations included 18 different combinations of weather variables

May Model Validation (2)  Also estimate “synthetic” event-day models  Test significance of coefficients on variables for event-like non-event days  Coefficients that are not statistically significant indicate that models do not falsely estimate load impacts on non-event days  Examine sensitivity of estimated hourly load impacts on actual event days across 18 alternative specifications  Compare predicted to actual loads on event-like days

May Model Validation (3)  Findings from model validation:  Synthetic event tests do not find significant “false” load impact estimates  Little sensitivity of estimated load impacts across the tested specifications  Models predict well on event-like days

May Sensitivity of Load Impact Estimates to Weather Variable Specification (PG&E)

May Actual and Predicted Loads – Average Event-Like Non-Event (SCE)

May CBP Ex-Post Load Impacts (2011 – 13) Typical Event – Average Event-Hour (MW)

May Ex-Ante Load Impact Simulation Process 1-in-2 and 1-in-10 Weather Conditions Simulate Hourly Reference Loads Estimate Customer-level Regression Coefficients Calculate % Load Impacts (Based on 3 Years of Ex- Post LI) Enrollment Forecasts Ex-Ante Load Impacts per Customer Aggregate Load Impacts

May Comparison of Previous Ex-Ante to Current Ex-Post and Ex-Ante [CBP]

Key Factors in Ex-Ante Changes  PG&E believes that aggregators with both CBP and AMP contracts have focused on achieving AMP commitments, leading to lower CBP load impacts  SCE anticipates movement of AMP-DA contract to CBP-DA, and shifting less responsive customer accounts from AMP-DO to CBP-DO  SDG&E results varied somewhat due to  unexpected changes in customer nominations,  changes in performance of a few large customers, and  changes in the mix of % load impacts in previous 3 years May

May Questions?  Contact – Steve Braithwait or Dan Hansen, Christensen Associates Energy Consulting Madison, Wisconsin   