一种参数快速收敛的多变量系统神经网络PID控制

Parameter-converged Rapidly Neural Network PID Control for Multivariable System

对多变量系统设计了神经网络ＰＩＤ控制，解决了多变量系统Ｐ、Ｉ、Ｄ参数难以整定的问题。基于对象模型，提出了一种新型神经网络控制器的训练方法，该方法用多步二次型性能指标函数去训练控制器的权值，从而提高了控制器参数的收敛速度和系统的响应性能，降低了各通道之间的耦合。

Neural network PID control is designed for multivariable system, which solutes the difficulty that P,I,D parameter for multivariable system is not easily ascertained. Based on the plant model, a novel training method of neural network controller is proposed. The method uses multi-step quadratic index function to train the weights of the controller; therefore, it improves the convergence speed of the controller parameter and the system response performance, and reduces the coupling of all channels. The theoretical analysis and simulation experiment illustrate the effectiveness of the technique.