Combining AI and Simulation to Assess Building Demand Response Potential at Scale

Assessing building demand response (DR) potentials typically involves modeling building thermodynamics and conducting simulations. However, this approach requires detailed parameters or operational data for each building and becomes impractical at the provincial and national levels. Here, we propose DPAAS, a data-driven building DR potential assessment frame-work that combines artificial intelligence (AI) technology with advanced building energy simulation techniques. The framework envisions using readily available multimodal data sources to assess DR potential at scale, with machine learning models at its core to map building parameters to DR potentials. To validate the critical machine learning component of this framework, we conduct numerical tests using the Comstock dataset. Results show that the ML models can predict DR potentials with median errors of less than 10% for load shedding and 14% for load shifting, demonstrating the feasibility of the framework's core mechanism. This validation of the ML approach, combined with established methods for building data collection and processing, suggests that DPAAS could potentially enable cost-effective DR potential assessment at scale.