Building demand-side control using thermal energy storage under uncertainty: An adaptive Multiple Model-based Predictive Control (MMPC) approach

This study investigates the demand-side control enhancement of TES though comparing the benchmark control strategies and newly-suggested adaptive MMPC designed to handle uncertainty in an adaptive fashion. Evaluations are performed through closed-loop simulations of an actual test building that is calibrated with real data and weather. While typical MPC centralizes computational burden (due to a provision for disturbances) on optimization through a single model, the adaptive MMPC distributes such burden (with less complexity) to multiple local models and optimizations in advance, then the online supervisory controller selects or interpolates the most adequate local models and control policies for the current conditions, thereby provides an effective global control policy for the entire operation regime. Building Energy Model (BEM) is used to construct local models for the adaptive MMPC, which enables more semantically feasible and acceptable model calibrations through which practitioners would obtain more model fidelity. This approach not only alleviates real time computation load, but also still achieves the desired performance. Evaluation results show that the adaptive MMPC outperforms the storage priority control and also ensures a near-optimal performance in load shifting under various uncertainty situations, including depreciation scenarios and unmeasured disturbance scenarios.