Dynamic dispatch of an integrated energy system based on deep reinforcement learning in an uncertain environment
DOI：DOI: 10.19783/j.cnki.pspc.211685
Key Words:integrated energy system  dynamic dispatch  uncertainties  deep reinforcement learning  proximal policy optimization
Author NameAffiliation
LIN Weishan 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
WANG Xiaojun 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
SUN Qingkai 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
LIU Zhao 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
HE Jinghan 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
PU Tianjiao 1. School of Electrical Engineering, Beijing Jiaotong University, Beijing 100044, China
2. China Electric Power Research Institute Co., Ltd., Beijing 100192, China 
Hits: 3203
Download times: 844
Abstract:As the uncertainties of intermittent energy and load in the integrated energy system gradually increase, traditional dispatch methods are limited to fixed physical models and parameter settings that can hardly respond to the random fluctuations in the dynamic system with source-load. In this paper, a deep reinforcement learning-based dynamic dispatch method for the integrated energy system is proposed to address this problem. First, a data-driven deep reinforcement learning model is constructed for the integrated energy system. Through the continuous interaction between agent and integrated energy system, the dispatch strategies are learned adaptively to reduce dependence on the physical models. Secondly, the variations of source-load uncertainties are characterized by adding random disturbances. Pivotal aspects such as state spaces, action spaces, reward mechanisms and the training process of the deep reinforcement learning model are improved according to the characteristics of uncertainties. Then a proximal policy optimization algorithm is used to solve the problem, and the dynamic dispatch decisions of the integrated energy system are realized. Finally, simulation results verify the feasibility and effectiveness of the proposed method over different time scales and in uncertain environments. This work is supported by the National Natural Science Foundation of China (No. 51977005).
View Full Text  View/Add Comment  Download reader