基于深度确定性策略梯度算法的可再生能源大规模制氢系统能量调度  

Energy Scheduling of Renewable Energy Large-scale Hydrogen Production System Based on Deep Deterministic Strategy Gradient Algorithm

作　　者：梁涛 刘伟 曹欣[2] 谭建鑫[2] 孙鹤旭 LIANG Tao;LIU Wei;CAO Xin;TAN Jianxin;SUN Hexu(School of Artificial Intelligence,Hebei University of Technology,Hongqiao District,Tianjin 300130,China;Hebei Jiantou New Energy Co.,Ltd.,Shijiazhuang 050051,Hebei Province,China;School of Electrical Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,Hebei Province,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津市红桥区300130 [2]河北建投新能源有限公司,河北省石家庄市050051 [3]河北科技大学电气工程学院,河北省石家庄市050018

出　　处：《电网技术》2025年第4期1413-1425,I0030,共14页Power System Technology

基　　金：河北省科技支撑计划资助项目(E2024202051,244C4301D,22314601D)。

摘　　要：为了实现可再生能源充分利用、减少整流和并网等设备的投资成本、降低电解水制氢的成本,实现可再生能源大规模制氢。该文建立了离网型可再生能源大规模制氢系统(renewable energy-based hydrogen production system,H2-RES),并以H2-RES的能量管理为研究对象,优化目标为系统的经济性和安全性,首先搭建了H2-RES系统仿真环境,并给出了控制策略方法。然后提出一种基于深度确定性策略梯度(deep deterministic policy gradient,DDPG)算法的智能能量调度策略,采用DDPG算法进行长期大量训练学习得到的智能体可以实现智能化的实时在线能量调度,通过与深度Q网络(deep Q network,DQN)、粒子群优化(particle swarm optimization,PSO)、传统控制策略方法从经济性与安全性进行对比,验证了DDPG算法应用于H2-RES的能量管理可以获得更高的经济效益,消纳可再生资源能力更强,且可以保证系统的安全运行,具有较强的学术意义和工程价值。In order to realize the full utilization of renewable energy,reduce the investment costs of the rectifying and grid-connected equipment,lower the costs of the hydrogen production from the electrolytic water,and realize the large-scale hydrogen production from the renewable energy,this paper establishes an off-grid renewable energy large-scale hydrogen production system(H2-RES),Taking the energy management of H2-RES as the research object,the optimization objective is set as the economy and security of the system.First,the simulation environment of the H2-RES system is built,and the control strategy is given.Then,an intelligent energy scheduling strategy based on the Deep Deterministic Policy Gradient(DDPG)algorithm is proposed.The agent obtained through a long-term and massive training and learning with the DDPG algorithm may realize the intelligent real-time online energy scheduling.By comparing with the Deep Q Network(DQN),the PSO and the traditional control strategy in terms of economy and security,it is verified that the application of the DDPG algorithm to the energy management of the H2-RES is able to obtain higher economic benefits,to absorb more renewable resources,and to ensure the security of the system,having stronger academic significance and engineering value.

关 键 词：可再生能源 大规模制氢 孤岛 深度强化学习 

分 类 号：TM721[电气工程—电力系统及自动化]