Reinforcement learning-based demand response strategy for thermal energy storage air-conditioning system considering room temperature and humidity setpoints

Abstract

Heating, ventilation, and air-conditioning systems significantly influence demand response (DR) actions. In the literature, many studies have investigated the reduction of sensible loads through DR events by increasing temperature setpoints while satisfying indoor comfort levels. However, less consideration has been given to latent loads. Therefore, proximal policy optimization reinforcement learning (RL) algorithms are adopted to simultaneously control indoor temperature and humidity (T&H) setpoints to reduce sensible and latent loads. An EnergyPlus-Python joint simulation platform was created for the temperature-humidity independent control system. DR strategies based on RL, active thermal energy storage, and time-of-use electricity prices are formulated to find the optimal indoor T&H setpoints, considering environmental constraints, comfort levels, and energy consumption. The results showed that in the optimal experimental strategies implemented based on negative feedback regulation, savings of 29.15 % on electricity consumption were achieved during DR periods and 46.11 % savings on the daily operating costs. Moreover, a simulation strategy based on feed-forward regulation was developed and found to reduce electricity consumption by 6.92 % during DR periods. Compared with a fixed T&H setpoint, the developed strategy meets the preferences of different groups of people for thermal comfort and increases the system's flexibility.