Hybrid MPC for Building Heat Pump Demand Flexibility under Unmeasured Disturbances

Implementing advanced controls like Model Predictive Control (MPC) to unlock the grid flexibility of small- and medium-sized commercial buildings (SMCBs) is often prohibitively expensive due to the extensive sensors required to measure internal heat gains. This paper introduces a Hybrid MPC framework that overcomes this challenge without additional sensors by integrating a physics-informed gray-box model with a neural network (NN) that forecasts these unmeasured disturbances. Evaluated in a multi-building case study, the Hybrid MPC achieves substantial load shifting and peak demand reduction, with performance nearly matching an ideal controller with perfect foresight.