Data-Driven Predictive Control of Buildings; a Regression Based Approach

Abstract

In this paper, we present a data-driven predictive control (DDPC) strategy suitable for (general bilinear) building energy systems, which only relies on historical measurements. The introduced control technique operates iteratively in a receding horizon scheme. During the operation of DDPC, using available historical building data which are collected during the normal operation of building, the dynamics of building are approximated and estimated. Therefore, the system dynamics are decomposed into two stable subsystems, one describing the thermal dynamics of the mass of building and the other one describing the dynamics of the temperature of rooms leading to a linear infinite-dimensional interconnected system. Linear regression is used to estimate a finite approximation thereof. Employing the approximate model, a cost is minimized in each iteration. The performance of the introduced data-driven control method is numerically verified for a validated building simulation environment.