AlphaShed: A scalable load flexibility model for shedding potential in commercial HVAC systems

Abstract

Technologies that enable the demand flexibility (DF) in building loads have been identified as a key advancements to support the reliable operation of the electric grid. The concept of grid-interactive efficiency buildings (GEBs) envisions building loads actively controlling power consumption in alignment with grid services. Commercial HVAC systems, through load shedding, hold a significant portion of this resource. Due to the strict requirements from reliability-based grid services, this load shedding potentially needs to be predicted and communicated. However, due to the heterogeneity of these systems, developing a scalable and accurate model to meet this need is challenging. To this end, we propose AlphaShed: a load flexibility model that is based on the internal state of the HVAC system to estimate the magnitude of shedding events. The proposed feature, thermal unloading potential, contains information about the thermal balance in the zones as well as the allowed operational range of mechanical components. To test the proposed model we generate synthetic events in simulation for four standard building models across different weather scenarios with 800 events per instance. Afterwards, we fit a regression model predicting load shedding potential. We compare the proposed approach with a common model that leverages outdoor air temperature and demonstrate a 49%-88% reduction in errors when predicting the load shedding potential. Furthermore, we extend the validation with real-world experiments, showing the model can predict load shed events with high accuracy.