基于多智体强化学习的多风氢系统联合优化运行  

Joint Optimal Operation of Multi Wind-Hydrogen System Based on Multi-Agent Reinforcement Learning

作　　者：刘建树 江岳文[1] LIU Jianshu;JIANG Yuewen(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,Fujian Province,China)

机构地区：[1]福州大学电气工程与自动化学院,福建省福州市350108

出　　处：《现代电力》2022年第4期431-440,I0005,I0006,I0007,共13页Modern Electric Power

基　　金：国家自然科学基金项目(51707040)。

摘　　要：针对多风氢系统联合运行问题,提出一种基于多智体强化学习的多风氢系统联合优化运行方法,使得多风氢系统在有效消纳风电的同时实现联合收益最大化。首先,考虑风电场与制氢加氢站两者间通过合约交易方式联合运行,分别构建各自的运行模型;其次,以多风氢系统联合运行收益最大化为目标建立联合优化运行模型;再者,针对多风氢系统多决策变量导致的维数灾难问题,将多智体引入到强化学习中并采取多决策更新方法加速算法收敛;最后,通过算例仿真验证所提模型的合理性以及方法的可行性。In allusion to the joint operation of multi wind-hydrogen system,based on multi-agent reinforcement learning a multi wind-hydrogen system joint optimization operation method was proposed to make the multi wind-hydrogen system enable to accommodate wind power effectively and meanwhile to maximize the joint revenue.Firstly,considering the joint operation of wind farm,hydrogen generation and hydrogenation station in the manner of contract transaction,respective operation models for them were constructed.Secondly,taking the maximized joint operation revenue of multi wind-hydrogen system as the objective,a joint optimization operation model was established.Thirdly,to cope with the dimension disaster caused by multi decision variables of multi wind-hydrogen system,the multi agent was led into the reinforcement learning and the method of multi-decision update was adopted to speed up the algorithm convergence.Finally,the reasonableness of the established model and the feasibility of the adopted method are verified by simulation example.

关 键 词：多风氢系统 维数灾难 多智体 强化学习 多决策更新 

分 类 号：TM73[电气工程—电力系统及自动化]