AMI Billing Regression Study

Southern California Edison (on behalf of the four California investor-owned utilities) hired Evergreen Economics to conduct a study of how traditional billing regression analysis tools could be adapted for use with advanced metering infrastructure (AMI) data. Correctly understanding and leveraging the great wealth of information provided by AMI data (in addition to developing methods for systematically processing very large amounts of customer billing data) can revolutionize how energy efficiency programs are evaluated.

The Evergreen study presents a new approach—the AMI Customer Segmentation (AMICS) model—that allows savings estimates to be tailored more closely to individual customer characteristics. This is accomplished by first grouping customer consumption data into different categories based on energy use and weather conditions. Separate billing regression models (patterned after the random coefficients model specification) are then estimated for each usage/weather category, which allows for separate load shape predictions for very specific customer types.

In this study, the AMICS model specification was tested using data from two HVAC efficiency programs in California: Southern California Edison’s HVAC Quality Installation (QI) Program and Pacific Gas and Electric’s HVAC Quality Maintenance (QM) Program. Both of these programs had samples of over 1,000 customers and involved analyzing AMI billing data in 1-hour increments covering multiple years.

Using the AMI data from both programs, average daily load shapes were calculated for specific day types (weekday, weekend, seasonal) and used to estimate energy savings. When estimated load shapes were compared against a holdout sample of customers, the AMICS model performed extremely well; load shape predictions were within 1 percent of the actual load shapes for the holdout sample. Energy savings estimates for these programs ranged from 4 percent to 7 percent of annual energy use for the QM and QI programs respectively, which was consistent with the original savings expectations for these programs.

In addition to producing accurate impact estimates, the automated categorization and AMICS modeling processes developed by Evergreen allow for separate savings estimates and load shapes to be developed easily for a variety of different situations (e.g., time of day, day types, and seasons). The AMICS model also provides an opportunity to develop customer-specific predictions of energy use and potential savings associated with various efficiency programs, thus empowering utilities to target the most beneficial programs to each customer.