Data-driven Predictive Control of Office Temperatures Based on Multiple Linear Regression Model

The paper proposes a data-based approach in control of building temperature for the case of a single floor of an office building. Several obtained datasets are exploited to identify a mathematical model of building thermal dynamics. Input-output data mapping is performed by multiple linear regression to identify a linear model in form of a vector of coefficients related to historical variables of temperatures of the considered and neighboring offices, external variables and user behavior as disturbances. The model is exploited for temperature setpoint tracking optimal problem formulation and put as trade-off of comfort and energy efficiency. Such data predictive control approach is implemented to a case study and realistic simulations are performed with real data, proving the soundness of the approach. The concept is motivated by easy replication of both hardware and software to be efficiently implemented in the buildings, driven by time-effectiveness of the proposed structure and identification method of the model.