A Neural Network-based Model Predictive Control Approach for Buildings Comfort Management

This paper proposes a model predictive control (MPC) approach incorporated with machine learning to control the energy consumption and occupants' comfort (thermal and visual comfort) in a smart building. Neural networks (NN)s are developed to learn and predict the building's comfort specifications, environmental conditions, and power consumption. Based on the predicted data, MPC provides optimal control inputs for the thermal and lighting systems to achieve the desired performance. In contrast to the existing building control frameworks, our proposed learning-based control method incorporates the occupant-related parameters in the control loop, which enhances the prediction accuracy and control performance. Our proposed learning-based MPC approach is implemented on a building, simulated in EnergyPlus software, and its performance is compared with that of a model-based building control framework. From the simulation results, our control method performs significantly better than the conventional MPC in maintaining residents' comfort and reducing energy consumption.