3基于多智能体ceq学习的智能发电控制器设计-电测与仪表.doc

基于CEQ(()强化学习算法的微电网智能发电控制* 唐捷1，张泽宇2，程乐峰2，张孝顺2，余涛2 （1. 韶关供电局 广东 韶关 512026；2.华南理工大学电力学院 广州 510640） 摘要：针对微电网孤岛运行模式下新能源发电强随机性导致的系统频率波动，提出了基于多智能体相关均衡强化学习(Correlated Equilibrium Q(λ), CEQ(())的微电网智能发电控制方法。在所搭建含有光伏发电、风力发电、小水电、微型燃气轮机和飞轮储能的微电网负荷频率控制(Load frequency Control, LFC)模型基础上，以频率偏差作为状态输入，提出了一种微电网孤岛运行模式下的CEQ(()智能发电控制器。仿真结果显示，与PI控制、单智能体R(()控制相比，CEQ(()控制器具有更好的在线学习能力，能显著增强孤岛微电网的鲁棒性和适应性，有效提高了频率的考核合格率。 关键词：多智能体；相关均衡；强化学习；微电网；智能发电控制 中图分类号：TM76 文献标识码：B 文章编号：1001–1390（201）0–0000–00 Smart generation control for micro-grids based on correlated equilibrium Q(() learning algorithm Tang Jie1, Zhang Zeyu2, Cheng Lefeng2, Zhang Xiaoshun2, Yu Tao2 (1. Shaoguan Power Supply Bureau of Guangdong Power Grid Co., Ltd. , Shaoguan 512026, Guangdong, China. 2. College of Electric Power, South China University of Technology, Guangzhou 510640, China) Abstract：In view of addressing frequency fluctuation problem of the islanded micro-grid caused by strong randomness of renewable energy, this paper proposes a smart generation control method of micro-grid based on multi-agent correlated equilibrium Q(λ). After establishing the micro-grid load frequency control (LFC) model containing photovoltaic generation, wind generation, small hydro unit, micro-gas turbine and flywheel energy storage, which takes frequency deviation as the input, a smart generation controller based on correlated equilibrium Q(λ) has been proposed. Simulation analysis, compared with PI controller, R(λ) learning controller, shows that correlated equilibrium controller has better online learning ability, and enhances robustness and adaptability of islanded micro-grid remarkably and improves the pass rate of frequency assessment effectively. Keywords：multi-agent, correlated equilibrium, reinforcement learning, micro-grid, smart generation control0　引 言 自1998年CERTS提出微电网概念至今，各国学者对微电网展开了深入的研究[1-2]。微电网中包含大量的可再生能源，其出力受环境变化影响很大。这种随机性和间歇性给微电网的控制带来了极大的挑战。尤其当微电网处于孤岛运行模式下，其频率和(电压失去了大电网的支撑，受可再生能源波动的影响更为严重。 由于传统PI控制器难以满足微电网复杂的运行