With the growing global emphasis on energy conservation and emission reduction, optimizing integrated energy systems (IES) has become essential for enhancing energy efficiency and lowering pollutant emissions. However, traditional methods often struggle with high-dimensional continuous action spaces due to overestimation bias, which affects strategy stability. This paper investigates a multiobjective approach for optimizing electricity–gas–heat integrated energy systems (EGH-IES) using a deep reinforcement learning approach. Firstly, we consider wind power absorption and pollutant emissions, and applying the operation mechanisms and constraints of devices in EGH-IES to construct optimization model involving operating costs and environmental factors. Moreover, the optimization problem is reformulated as an interactive reinforcement learning process between the agent and its environment. Then, dual-critic networks and delayed policy updates are introduced to reduce overestimation errors, and the twin delayed deep deterministic policy gradient (TD3) algorithm is used to optimize the device output power. Finally, simulation results show that compared with deep deterministic policy gradient (DDPG) and proximal policy optimization (PPO), the method presented increased the reward function value by 4.87% and 11.36%, respectively. By comparing power balance results across seasons, the system could accurately identify seasonal features and adjust device strategies dynamically.
Li et al. (Thu,) studied this question.