Data Efficient Energy Disaggregation with Behind-the-Meter Energy Resources

The deployment of behind-the-meter energy resources is set to increase in the next decade. Yet, utilities are not ready to safely manage these resources owing to the lack of visibility into behind-the-meter installations. To address this problem, we propose an unsupervised method to decompose a customer’s net load to key components, namely aggregate home load, solar generation, and battery charge/discharge activities, without the knowledge of their deployment characteristics and control strategy. The proposed method takes advantage of proxy measurements from one separately metered PV system in the proximity of the target customer besides solar generation traces synthesized using a physical PV model. Using smart meter and solar inverter readings from 140 homes, we evaluate this method in two scenarios where the customer installed a PV system and where this PV system is connected to a battery. Assuming the availability of data from one real PV system and three synthetic proxies, we show that our method reduces the RMSE of solar generation estimates by 31.59% and 21.87% on average over two baselines when battery is not installed. In the presence of a behind-the-meter battery, we show that it reduces the RMSE of solar generation estimates by more than 8.67% over the state-of-the-art method that accounts for the battery operation