Real-time Optimal Energy Management of Combined Cooling, Heating and Power systems Based on Knowledge conlusion

Combined Cooling, Heating and Power(CCHP) system is an important way to improve the efficiency of energy utilization and wind power consumption. In this case, increasing the operational flexibility of CCHP systems can help increase renewable energy consumption and reduce carbon emissions. So, it is necessary to provide real-time optimal energy management(OEM) for CCHP systems. In this paper. Firstly, a new training method is proposed for the LightGBM to improve the accuracy of real-time wind power prediction. In addition, to solve the nonlinear OEM of CCHP in real-time, a novel deep reinforcement learning algorithm-Soft Actor-Critic, is used in this paper. By combining the mechanism of CCHP with the reward function through knowledge fusion, the non-convex models can be translated into Markov Decision Processes(MDP), and then according to the Bellman optimal equation, the real-time optimized operation of the CCHP system is realized. Finally, the effectiveness of the proposed policy is verified by the actual operation data of a CCHP system in Northwest China.