综合智慧能源2025，Vol.47Issue(1)：18-25,8.DOI:10.3969/j.issn.2097-0706.2025.01.003
基于深度强化学习的风电场功率多变量综合优化控制
Multivariable integrated power control optimization of wind farms based on deep reinforcement learning
摘要
Abstract
China has proposed the"dual carbon"strategy,aiming to build a new power system with renewable energy as the primary component.Based on real wind farm data,optimization control strategies were proposed to improve the wind farm's output power,thereby further enhancing wind energy utilization.The wake effects between wind turbines were the main focus,and a wind farm power multivariable optimization control strategy based on model-free deep reinforcement learning(DRL)was proposed.The strategy employed the Proximal Policy Optimization(PPO)algorithm to optimize multiple variables,including the yaw angle,tilt angle,blade pitch angle,and tip speed ratio(TSR)in a dynamic wind farm.Through intelligent agents learning from the data generated by the agent during operation,an optimal control strategy was obtained,overcoming the limitations of traditional mathematical optimization methods.Simulation results showed that,compared to existing wind turbine control algorithms,the model-free DRL-based multivariable optimization control strategy significantly improved computational efficiency,reduced the difficulty of parameter optimization,and optimized the direction and strength of the wake.The optimized average output power was increased by 37.08%.关键词
风电场/深度强化学习/尾流控制/多变量控制/功率优化/近端策略优化/"双碳"目标/新型电力系统Key words
wind farm/deep reinforcement learning/wake control/multivariable control/power optimization/PPO/"dual carbon"strategy/new power system分类
能源科技引用本文复制引用
张华钦,刘伟,王慧,李雷孝,莎仁高娃..基于深度强化学习的风电场功率多变量综合优化控制[J].综合智慧能源,2025,47(1):18-25,8.基金项目
国家自然科学基金项目(62362055) （62362055）
内蒙古自治区气象局科技创新项目(nmqxkjcx202312,nmqxkjcx202435)National Natural Science Foundation of China(62362055) （nmqxkjcx202312,nmqxkjcx202435）
Scientific and Technological Innovation Project of Meteorological Bureau of Inner Mongolia Autonomous Region(nmqxkjcx202312,nmqxkjcx202435) （nmqxkjcx202312,nmqxkjcx202435）
