Low-carbon Economic Dispatch of Electric-thermal Coupling System Based on Twin Delayed DDPG Reinforcement Learning

For the electric-thermal coupling system with renewable energy access, a reinforcement learning method is proposed for low-carbon economic dispatch of electric-thermal coupling systems. Firstly, a low-carbon economic dispatch model of the electric-thermal coupling system is established with both the economy and carbon emissions taken into account. The low-carbon economic dispatch process of the electric-thermal coupling system containing renewable energy is subsequently transformed into a Markov decision process (MDP). With the aim of minimizing both the economy and carbon emissions, a multi-objective reward function is designed by combining the penalty constraint mechanism. Additionally, based on the improved algorithm of deep deterministic policy gradient (DDPG), a twin delayed DDPG algorithm is utilized to train reinforcement learning agents interactively. Finally, the numerical result demonstrates that the agent trained by the proposed method can respond to the uncertainty of renewable energy and electric/thermal load in real time, enabling the optimization of the low-carbon economic scheduling for the electric-thermal coupling system containing renewable energy online.