An Improved TD3 Framework for CHP Economic Dispatch: Synergistic Optimization of Entropy Reward Mechanisms and Adaptive Noise

This study proposes AME-TD3, an enhanced deep reinforcement learning algorithm, to address the high computational complexity and poor adaptability of traditional methods in optimizing combined heat and power (CHP) systems. AME-TD3 integrates an entropy reward mechanism to balance exploration-exploitation and adaptive noise adjustment for dynamic system response, improving stability. A CHP economic dispatch model with power, CHP, and heat only units minimizes costs while meeting supply-demand constraints. Case studies (24/48-unit systems) show AME-TD3 outperforms TD3, DDPG, and particle swarm methods in minimum, average costs and standard deviation, demonstrating superior adaptability in large-scale dynamic environments. Results confirm its effectiveness in economic optimization, stability, and multi-constraint handling for integrated energy systems.