基于深度强化学习算法的风光互补可再生能源制氢系统调度方案  

Scheduling Scheme of Wind-solar Complementary Renewable Energy Hydrogen Production System Based on Deep Reinforcement Learning

作　　者：梁涛 孙博峰 谭建鑫 曹欣 孙鹤旭 LIANG Tao;SUN Bofeng;TAN Jianxin;CAO Xin;SUN Hexu(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300130,China;Hebei Construction&Investment Group Co.,Ltd.,Shijiazhuang 050051,China;School of Electrical Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300130 [2]河北建投能源投资股份有限公司,石家庄050051 [3]河北科技大学电气工程学院,石家庄050018

出　　处：《高电压技术》2023年第6期2264-2274,共11页High Voltage Engineering

基　　金：河北省科技支撑计划(19210108D,19214501D,20314501D,F2021202022)。

摘　　要：风光可再生能源制备“绿氢”是实现能源低碳化的重要途径,但风能、太阳能的波动性、间歇性等问题会使系统存在“弃风、弃光”现象。为解决该问题,构建了可再生能源并网制氢系统,针对传统CPLEX需要精准预测数据、基于状态控制法的监控策略控制效果不够理想的缺点,将协调控制转化为序列决策问题,采用深度强化学习连续近端策略优化算法进行解决。在发电量、负荷等多种因素变化的情况下,设计了适合解决可再生能源制氢系统调度问题的深度强化学习模型(renewable energy to hydrogen-proximal policy optimization,R2H-PPO),经过足够的训练后能够实现在线决策控制,并与日前控制方案和基于状态控制法的监控策略进行了对比,证明所采用方法避免了传统方案的不足,并能有效处理不同时刻、天气、季节的场景。结果证明了所提出的R2H-PPO方法的可行性和有效性。The preparation of“green hydrogen”from wind-solar renewable energy is an important way to achieve low carbon energy,but the volatility and intermittency of wind and solar energy will cause the phenomenon of“wind and solar abandonment”in the system.In order to solve this problem,a grid-connected hydrogen production system of renewable energy is constructed in this paper.In view of the shortcomings that traditional CPLEX requires accurate data prediction and the control effect of monitoring strategy based on state control method is not ideal,the coordinated control is transformed into a sequential decision problem,which is solved by deep reinforcement learning continuous proximal policy optimization algorithm.Under the variation of various factors such as electricity generation and load,a deep strengthening learning model(R2H-PPO)suitable for solving the scheduling problem of hydrogen production system from renewable energy sources is designed,thereby,online decision controls can be realized after sufficient training;moreover,the online decision controls are compared with day-ahead control scheme and monitoring strategy based on state control method.It is proved that the adopted method avoids the shortcomings of traditional scheme,and can effectively deal with different time,weather and season scenarios.The results show that the proposed R2H-PPO method is feasible and effective.

关 键 词：可再生能源 制氢系统 深度强化学习 近端策略优化 运行优化 R2H-PPO 

分 类 号：TQ116.2[化学工程—无机化工] TP18[自动化与计算机技术—控制理论与控制工程]