Data Analysis of commercial customers using the AMI Customer Segmentation (AMICS)

Conduct AMI data analysis of individual commercial customers using the AMI Customer Segmentation (AMICS) modeling approach. The California Public Utility Commission (CPUC) recently approved Southern California Edison’s (SCE) application for a High Opportunity Programs or Projects (HOPPs) HVAC program. However, this approval is contingent on a demonstration proving that the AMICS model’s ability to capture site level savings for individual customers. The AMICS model is one among many different models available for conducting Normalized Metered Energy Consumption (NMEC) M&V.

PROJECT GOAL – The goal of the study was to demonstrate that the AMI Customer Segmentation (AMICS) model is able to produce reliable savings estimates for individual non-residential customers, and is thereby able to meet California’s requirements of normalized metered energy consumption analysis. TECHNOLOGY DESCRIPTION – The AMICS model was originally developed for the purpose of estimating hourly energy savings for energy efficiency programs using AMI utility billing data. Evergreen Economics’ previous research demonstrated that the AMICS modeling approach produces similar results to a traditional fixed effects regression model at the program level, while providing valuable insights into the characteristics of customers and weather conditions that drive these savings. AMICS achieves this by segmenting the AMI data into a series of discrete analysis segments, each containing customers with similar energy usage patterns on days with similar weather conditions. To demonstrate the efficacy of the AMICS model approach, Evergreen Economics applied the AMICS model to a small sample (n=10) of SCE commercial HVAC program participants. For comparison, the same data were used in LBNL’s Temperature and Time of Week (TTOW) model, which has seen wider acceptance as an approach for estimating savings using AMI data. A key benefit of the AMICS model is avoiding over-reliance on ‘average day’ conditions. Models like TTOW essentially estimate the average load shape and then make a series of adjustments to that prediction depending on how the actual weather conditions differ from this average. The AMICS approach uses segmentation to produce a portfolio of load shapes and then compares each day in the post-period against similar days in the pre-period. PROJECT FINDINGS – This analysis did not find any significant differences in the prediction error between these two modeling approaches. We believe that both the AMICS and TTOW approaches are both well suited for AMI analysis of residential and commercial customers, and choosing one model over the other will not significantly affect the analytical results.