作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2010, Vol. 36 ›› Issue (1): 18-20,2. doi: 10.3969/j.issn.1000-3428.2010.01.007

• 博士论文 • 上一篇    下一篇

基于遗传算法和神经网络预测的再励学习

张华军,赵 金   

  1. (华中科技大学控制科学与工程系,武汉 430074)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2010-01-05 发布日期:2010-01-05

Reinforcement Learning Based on Genetic Algorithms and Neural Network Prediction

ZHANG Hua-jun, ZHAO Jin   

  1. (Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074)
  • Received:1900-01-01 Revised:1900-01-01 Online:2010-01-05 Published:2010-01-05

摘要: 提出一种基于遗传算法和神经网络预测法相结合的再励学习方法,利用遗传算法对全局进行最优解搜索,将进化过程中产生的数据用来训练神经网络预测器,当再励学习逼近最优解时,利用预测网络估计动作网络的参数、结构与系统响应之间的映射关系,用预测网络逼近最优解的能力引导遗传算法在局部向最优解快速逼近,以解决遗传算法局部振荡问题,从而实现快速学习的能力。将其应用于矢量控制交流电机的速度环控制器自学习中,仿真实验验证了该算法的有效性。

关键词: 再励学习, 遗传算法, 神经网络预测, 矢量控制

Abstract: This paper proposes a self-learning algorithms of neural networks based on Genetic Algorithms(GA) and neural network prediction. It uses GA to search the optimum resolution globally. The data generated during the evolutionary process is used to train the predictive networks. When the self-learning of networks converges on the adjacent domains of optimum resolution, it uses the prediction network to predict the map between network parameters and system response. The neural networks’ optimum resolution approximate property guides the GA to converge on the optimum resolution and solve the local oscillation problem of GA. Experimental result proves the fast convergence property of the proposed algorithm through design speed controller of asynchronous machine drive system

Key words: reinforcement learning, Genetic Algorithms(GA), neural networks prediction, vector control

中图分类号: