TA-23-C031 – Data-Driven Model to Capture Building Dynamics during Different Patterns of Power Outages PDF

In the U.S., electricity accounted for 41% of household end-use energy consumption in 2019. The demand side management of residential buildings, including heating, ventilation, and air conditioning (HVAC), home appliances such as refrigerators, washers/dryers, and lights, and behind-the-meter (BTM) distributed energy resources (DERs) such as PV and battery energy storage system (BESS), plays a critical role in alleviating the pressure on grid stability and reliability. Increasingly frequent extreme weather has highlighted the urgent need to systematically engage demand-side flexible loads and customer-generators such as BTM DERs in bridging the gap between power demand and supply. This gap is vitally important in extreme weather conditions, which may lead to an extensive power outage (e.g., Texas in February 2021). In some existing studies, an adaptive model predictive control (MPC) platform was established to assist residential buildings with PV and BESS in improving resilience under extreme weather conditions, where the ability to “accurately” predict building dynamics highly affects the performance of the MPC framework. Though most existing literature utilizes R2, CVRMSE, and MAE to evaluate the model performance, models with a good R2, CVRMSE, and MAE do not necessarily perform well when deployed in the MPC framework, especially for different power outage cases. This paper proposes a data-driven model to better capture the buildings’ dynamics during power outages, including HVAC power consumption, domestic hot water (DHW) power consumption, zone temperature, and DHW tank temperature. A detailed data mining flow work is presented. Furthermore, the performance of the proposed data-driven model is evaluated in a resilience-oriented MPC framework with real-world weather data for extreme cold and heat. Some preliminary results show that using this proposed data-driven model, the MPC framework’s control output oscillation phenomenon is significantly improved. It is concluded that the MPC controller’s performance is highly correlated with the model’s ability to capture the building dynamic’s changing trend. Scores from traditional performance evaluation metrics for forecast models, including R2, CVRMSE, and MAE, cannot guarantee the MPC controller’s performance. The capability to better capture building dynamics in terms of variation and trend in frequent switching between power-off and power-on states helps improve the MPC controller’s performance.