基于强化学习的微电网能量调度优化策略研究  

Research on Energy Scheduling Optimization Strategy of Micro-grid Based on Reinforcement Learning

作　　者：吴利刚 张梁 周倩 李承子 WU Li-gang;ZHANG Liang;ZHOU Qian;LI Cheng-zi(School of Mechanical and Electrical Engineering,Shanxi Datong University,Datong 037003,China;School of Coal Engineering,Shanxi Datong University,Datong 037003,China;School of Business,Shanxi Datong University,Datong 037003,China;State Grid Shanxi Electric Power Company,Taiyuan 030000,China)

机构地区：[1]山西大同大学机电工程学院,山西大同037003 [2]山西大同大学煤炭工程学院,山西大同037003 [3]山西大同大学商学院,山西大同037003 [4]国家电网山西省电力公司,山西太原030000

出　　处：《控制工程》2022年第7期1162-1172,共11页Control Engineering of China

基　　金：山西省高等学校科技创新项目(2021L391);安徽省智能机器人信息融合与控制工程实验室开放课题项目(IFCIR2020006);大同市软科学研究计划项目(2020159);山西大同大学科学研究项目(2020YGZX014,2021YGZX27)。

摘　　要：针对需求侧动态电价响应下的城市微电网能量调度问题进行研究。首先,基于强化学习算法,将分级电力市场能量供需问题建模为离散-有限的马尔可夫决策过程;然后,通过Q-Learning贪婪算法求解不同类型能耗区域的最佳电力能耗与电力价格。实验结果表明,通过建立能源调度中心实现了工业园区、商业中心、住宅区域3个典型能耗区域中电力资源1.31%、5.16%、6.63%的缩减,在保证电网可靠性的前提下,提高了电力资源的合理利用率并降低了消费者的购电成本。The energy scheduling problem of urban micro-grid under dynamic demand-side tariff response is studied.First,based on the reinforcement learning algorithm,the energy supply and demand problem of hierarchical power market is modeled as a discrete-finite Markov decision process.Then,the optimal power energy consumption and power price for different types of energy consumption regions are solved by the Q-Learning greedy algorithm.The experimental results show that the energy dispatch center achieves 1.31%,5.16% and 6.63% curtailment of power resources in three typical energy consumption regions of industrial parks,commercial centers and residential areas,which improves the reasonable utilization of electricity resources and reduces the cost of electricity purchase for consumers while ensuring the reliability of the grid.

关 键 词：强化学习 智能微电网 动态价格 能源管理 Q-LEARNING 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]