基于最小资源分配网络的自适应PID控制

Adaptive neuron PID control based on minimum resource allocation network

针对常规径向基函数神经网络(radial basis function neural network,RBFNN)的逼近性能对网络结构和初始参数依赖性强的问题,采用最小资源分配网络进行改进,并与单神经元PID控制相结合,提出了一种基于最小资源分配网络的自适应PID控制方法。该方法利用最小资源分配网络动态构建RBFNN,实现RBFNN结构和参数的在线优化,并用该RBFNN辨识对象的离散模型,然后由单神经元PID控制器完成PID参数的自适应整定。仿真结果表明,该方法中PID参数能够很好地适应系统输入信号的变化,对非线性系统控制效果较为理想。

To solve the problem that the approximation performance of conventional RBFNN had a strong dependence on net- work structure and initial parameters, this paper used minimal resource allocation network to dynamically construct RBFNN, and combined the RBFNN with single neuron PID control, proposed an adaptive neuron PID control method based on minimum resource allocation network. Firstly, the method used the minimum resource allocation network to optimize the structure and parameters of RBFNN on line, and identified the discrete model of the controlled object by the RBFNN. Then it used single neuron PID controller to tune the controller parameters. The simulation result indicates that the parameters of PID controller in this method can adapt well with the changes of the system input signal, and has a more ideal control effect on nonlinear sys- tem.