Combining Data-driven and Physics-based Process Models for Hybrid Model Predictive Control of Building Energy Systems

Abstract

Model predictive control is well suited to control building energy systems efficiently. However, it still lacks commercial relevance due to the high modeling effort. This article presents a methodology to reduce the modeling effort by combining data-driven and physics-based process models in a hybrid MPC scheme. Data-driven models like artificial neural networks are generally nonconvex and nonlinear. Thus, using such models results in a nonlinear, nonconvex optimization problem. We present a workflow to efficiently solve the resulting optimization problem with gradient-based solvers using the algorithmic differentiation tool CasADi. The developed workflow is applied to an exemplary building energy system to implement an economic, hybrid model predictive controller. Simulation results confirm the high potential of the proposed methodology by realizing a cost-effective operation of the controlled system.