Data-Driven Predictive Control of Bilinear HVAC Dynamics—An Experimental Case Study

Abstract

Buildings are responsible for around 40% of the global energy demand. In order to effectively reduce the high energy consumption of HVAC systems while maintaining comfortable indoor climate, tailored control schemes are promising. Since the derivation of physical models of individual HVAC systems is time consuming, data-driven methods are a promising alternative. This letter proposes a framework for data-driven predictive control of HVAC system with bilinear system dynamics, which compensates for prediction errors via constraint adaptation through a bias term. The proposed scheme combines an extension of Willems’ fundamental lemma to bilinear systems with the consideration of multiple data-sets. To evaluate the efficacy of the data-driven control scheme, an experimental case study is performed under realistic conditions. In comparison with an existing simple control scheme, our results demonstrate energy efficient operation and successful compensation of prediction errors.