Key variables in determining energy shaving capacity of buildings during demand response events

In the US, the increasing electricity demand lays massive pressure on national grids. Demand response (DR) programs provide an economical way to avoid electrical blackouts by incentivizing end-consumers to reduce their demand during peak hours and emergencies. However, the estimation of the power demand shaving capacity (PSC) (i.e., the amount of power demand that can be curtailed) of buildings is often inaccurate in the current practice because it only relies on oversimplified building design specifications. Moreover, existing estimation models of PSC either require detailed information (e.g., thermal resistance value) that are not readily available and hard to acquire, or that are building/system-specific and hard to scale up. In this study, the authors implemented the state-of-the-art feature explanation approach to identify key variables by evaluating the contribution of all related variables extracted from previous DR research efforts and current practices of DR programs. By identifying key variables, the authors attempted to find the sweet spot of PSC estimation models that can ensure accuracy, generality, and scalability. The proposed data-driven PSC estimation model using only key variables (47% reduction from all identified information items) and trained using 28 different buildings showed 81% better performance as compared with the benchmark of the current practice.