Predictive and probabilistic modelling using machine learning for building indoor climate control

For the last few decades, thermal comfort has been considered an aspect of sustainable building evaluation methods and tools. However, estimating the indoor air temperature of buildings is a complicated task due to the nonlinear behaviour of heating, ventilation and air conditioning systems combined with complex dynamics characterized by the time-varying environment with disturbances. This issue can be alleviated by modelling the building dynamics using Gaussian processes since it also measures the uncertainty bounds. The main focus of this paper is designing a predictive and probabilistic room temperature model of buildings using Gaussian processes and incorporating it into model predictive control to minimize energy consumption and provide thermal comfort satisfaction. We exploited the Gaussian processes’ full probabilistic capabilities as the mean prediction for the room temperature model and used the model uncertainty in the objective function not to lose the desired performance and to design a robust control scheme. We illustrated the potentials of the proposed method in a numerical example with simulation results