Performance based thermal comfort control (PTCC) using deep reinforcement learning for space cooling

Abstract

With the recent increase in energy consumption in buildings, energy-saving strategies in buildings have become a priority in the energy policies of many countries. Therefore, many recent research studies have emphasized the advanced control methods to attain comfortable thermal conditions while minimizing the energy consumption in buildings. A new approach of thermal comfort control for space cooling/heating system is needed to reflect the changing indoor environment information in real time, and to control various factors (e.g., humidity, air velocity, etc.) that affect not only the temperature but also the thermal comfort.In this study, we propose the Gaussian process regression (GPR) for real-time thermal comfort prediction, a data-driven approach. These data-driven approaches will enable the monitoring of occupants and thermal comfort conditions based on real-time data and situational awareness. Then, based on the thermal comfort performance (PMV) prediction results obtained using the GPR, we investigated control methods involving the integration of systems, i.e., a variable refrigerant flow (VRF) system and a humidifier, instead of using simple set-temperature control for space cooling. For this purpose, deep Q-learning, which is an reinforcement learning method, was employed to derive the VRF and humidification integrated control methods. During zone operation, this algorithm learned an effective control policy based on rewards (thermal comfort and energy consumption) without relying on a thermal dynamics model. Moreover, by comparing the thermal comfort and energy consumption results with those obtained using fixed set-point (rule-based) control and performance-based comfort control for cooling, the efficiency of the proposed performance-based thermal comfort control (PTCC) was evaluated.As a results, it was found that PTCC yielded the optimal control action value that minimized the energy consumption while satisfying the thermal comfort conditions. In addition, applying the proposed PTCC strategy to cooling control could maintain the required performance level of thermal comfort by reflecting changing environmental conditions in real time, unlike the fixed set-point control.