Fast stochastic predictive control for building temperature regulation

This paper presents a nonlinear stochastic model predictive control (MPC) approach to building thermal temperature regulation. The control objective is to minimize energy consumption while bounding the probability of thermal comfort violations using prediction of weather and occupancy. We exploit the structure of the bilinear thermal network model and propose a partial closed-loop control scheme. This allows analytical computation of the predicted state variance matrices and easy reformulation of the stochastic MPC problem into a nonlinear program (NLP). We present a tailored sequential quadratic programming method to solve the NLP by exploiting its sparsity. Simulation results show good performance and computational tractability of the resulting scheme.