摘要
为克服常规神经网络只能离线训练的不足,提出了一种基于遗传算法的改进递归神经网络。首先从PI自整定控制器性能指标出发,阐述了递归神经网络原理及异步电机转速控制系统模型,说明了采用遗传算法来寻找卡尔曼滤波最佳参数的可行性。以异步电机控制系统为平台,借助MATLAB/Simulink软件进行了建模与仿真。仿真结果表明提出的基于遗传算法的改进神经网络获得的PI参数能有效减小系统超调,系统具有良好的动静态性能。
To overcome the disadvantage of conventional neutral network which is training only offline, an improved neutral network based on genetic algorithm is proposed. Starting from the performance index of self-tuning PI controllers, principle of recurrent neutral network and speed control system model of induction motor are elaborated. Then, the feasibility of by genetic algorithm to seak optimal constants of extended kalman filter is explained. Finally, modeling and simulation are performed by MATLAB/Simulink with control system of induction motors as platform. The results show that the PI parameters obtained by the proposed improved neutral network based on genetic algorithm can reduce the overshoot of system effectively and the system has good dynamic and static performance.
出处
《电气自动化》
2012年第6期3-5,共3页
Electrical Automation
关键词
遗传算法
递归神经网络
扩展卡尔曼滤波
PI
异步电机
genetic algorithm
recurrent neural network
extended kalman filter
PI
induction motor
