JAUE2020-061: Optimization of the operation of CCHP system with PV system based on deep reinforcement learning

Abstract

CCHP system is combined cooling, heating, and power, and it is important to minimize operating costs while meeting load demands. In this paper, an optimization model of the CCHP system considering the uncertainty of PV generation method is presented. Firstly, the K-means algorithm is used to cluster the full-year load data, and after clustering, all the hourly load data of each type are averaged, and then the cooling, heating, and power load of a typical load day is obtained. Secondly, the Q-learning method, which is a classical reinforcement learning method, is used to solve the optimization problem of typical days, and the state contains the generating power of the prime mover, cooling power of the absorption refrigerator, and the state of power storage. With the operating costs of the system as the environmental feedback, the model-free method can learn autonomously by changing the work power of the prime mover and absorption refrigerator. Finally, the given linear optimization result is compared with the reinforcement learning result, and the operating costs of the system operating strategy acquired through the reinforcement learning method are close to the optimal result on the typical days.