摘要
针对传统PID整定控制效果差且单纯神经网络整定存在参数学习和调整困难等问题,提出了一种基于改进模糊神经网络的PID参数整定方法。在该方法中,PID控制器的控制参数采用基于Mamdani模型的模糊神经网络进行自适应整定,模糊神经网络参数采用混沌遗传算法离线粗调和BP算法在线细调的方式进行学习和调整,仿真结果表明该整定策略动态响应快、误差控制精度高且网络中各节点及参数物理意义明确。最后分别从模糊规则数的变化及适应度函数的选取两方面提出两种优化方案,仿真结果表明增加模糊规则数或采用不同的适应度函数都有利于进一步减小控制误差。
Considering the poor effect in traditional PID tuning and parameters adjustment difficulties in neural networks, this paper proposed a new method of self-tuning of PID parameters based on improved fuzzy neural network. In this control method, the control parameters of PID controller were adaptively adjusted by fuzzy neural network based on Mamdani model, the param- eters of fuzzy neural network were optimized by the hybrid learning methods integrating the offline chaos genetic algorithm for coarse adjustment, with the online BP algorithm for precise adjustment, the simulation results show that the tuning strategy has fast dynamic response and high error control accuracy, every nodes and parameters have explicit physical meaning. Finally, it proposed two inproved scheme respectively from the number of fuzzy rules and the selection of fitness function, the simulation results show that the increase of fuzzy rules and the different fitness function is helpful to further reduce the control error.
出处
《计算机应用研究》
CSCD
北大核心
2016年第11期3358-3363,3368,共7页
Application Research of Computers
基金
上海市自然科学基金资助项目(12ZR1420700)
