A Data-Driven Thermal Sensation Model Based Predictive Controller for Indoor Thermal Comfort and Energy Optimization

Abstract

This paper proposes a model predictive controller (MPC) using a data-driven thermal sensation model for indoor thermal comfort and energy optimization. The uniqueness of this empirical thermal sensation model lies in that it uses feedback from occupants (occupant actual votes) to improve the accuracy of model prediction. We evaluated the performance of our controller by comparing it with other MPC controllers developed using the Predicted Mean Vote (PMV) model as thermal comfort index. The simulation results demonstrate that in general our controller achieves a comparable level of energy consumption and comfort while eases the computation demand posed by using the PMV model in the MPC formulation. It is also worth pointing out that since we assume that our controller receives occupant feedback (votes) on thermal comfort, we do not need to monitor the parameters such as relative humidity, air velocity, mean radiant temperature and occupant clothing level changes which are necessary in the computation of PMV index. Furthermore simulations show that in cases where occupants’ actual sensation votes might deviate from the PMV predictions (i.e., a bias associated with PMV), our controller has the potential to outperform the PMV based MPC controller by providing a better indoor thermal comfort.