人工情感Q学习的互联电网自动发电控制算法  

Artificial emotionnal Q-learning for automatic generation control of interconnected power grids

作　　者：殷林飞[1] 郑宝敏 余涛[1] YIN Lin-fei;ZHENG Bao-min;YU Tao(School of Electric Power, South China University of Technology, Guangzhou Guandong 510640, China)

机构地区：[1]华南理工大学电力学院,广东广州510640

出　　处：《控制理论与应用》2016年第12期1650-1657,共8页Control Theory & Applications

基　　金：国家自然科学基金项目(51177051;51477055);国家"973"计划项目(2013CB228205)资助~~

摘　　要：对互联电网中自动发电控制AGC中控制策略进行改进,设计了人工智能中的人工心理学和人工智能中的机器学习结合的控制策略.分别对Q学习算法和Q(λ)学习算法进行改进,设计了具有人工情感的智能体.提出了人工情感Q学习算法和人工情感Q(λ)学习算法.且将人工情感分别作用于Q学习算法和Q(λ)学习算法中的输出动作、学习率和奖励函数.最后在IEEE标准两区域和南方电网四区域的互联电网Simulink模型中进行数值仿真.绘制并统计了控制性能指标、区域控制误差和频率偏差的值.从仿真结果看,所提人工情感Q学习算法和人工情感Q(λ)学习算法控制效果优于原有Q学习算法、Q(λ)学习算法、R(λ)算法、Sarsa算法、Sarsa(λ)算法和PID控制算法,该数值仿真结果验证了所提算法的可行性和有效性.Artificial psychology and machine learning are combined in the automatic generation control strategy of interconnected power grids. An agent obtaining artificial emotion is designed, and the Q-learning and Q()-learning algorithms are improved by artificial emotion. The novel artificial emotional Q-learning and artificial emotional Q()-learning algorithms are proposed. The artificial emotion is respectively applied to the selection of output action, learning rate and reward function in Q-learning and Q()-learning, and then simulated on the standard IEEE two-area model and the China Southern Power Grid four-area model. The control performance standard, area control error and frequency deviation are figured. Simulation results verify the feasibility and effectiveness of the proposed algorithms and their superiority to the Q-learning, Q()-learning, R(), Sarsa, Sarsa() and PID algorithms.

关 键 词：人工情感 Q学习 Q()学习 自动发电控制 

分 类 号：TM76[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]