摘 要近年来，工业控制技术飞速发展，为了满足更高级别的工业控制要求，控制系统 的设计越来越复杂。工业生产中大多数的控制对象都是多输入多输出控制系统，并且 它们之间相互关联，因此，使用常规的单输入单输出控制系统就很难获得较好的控制 效果。在这种情况下，更多的针对多变量耦合控制系统的解耦控制算法不断涌现。83700

本文首先针对某控制系统中的双输入双输出系统模型设计对角阵解耦控制器并 在 Simulink 中搭建仿真模型。接着着重研究了 PID 神经网络解耦算法，以某一耦合 情况严重的 3 输入 3 输出系统为研究对象设计 PID 神经网络解耦算法，并提出了增加 动量项和增加神经元系数两种改进方法。为了进一步优化权值，本文最后提出了一种 带粒子群算法的 PID 神经网络解耦控制方法，并在 MATLAB 中进行仿真。

仿真结果表明:对角阵解耦算法面对不太复杂的被控系统时，可以较好地完成解 耦任务，获得较好的控制效果；经 PID 神经网络解耦后，系统的输出能够跟踪设定的 期望值，目标函数的误差值经过一段时间后趋近于零，但是仍存在稳态误差较大，调 节时间不够迅速等问题；增加动量项和神经元系数能够在一定程度上改善系统的动静 态性能；加入粒子群算法的 PID 神经元网络不管在稳态误差还是在调节时间等方面， 解耦效果都更加令人满意。

毕业论文关键词：PID 神经网络；多变量系统；粒子群算法；解耦控制

Abstract In recent years, the industrial control technology is developing very rapidly。In order to meet the requirements of a higher level of industrial control,The design of control system is becoming more and more complicated。 Most control objects in industrial production are multi-input and multi-output control system and they are related to each other very closely。Therefore, it is difficult to get a good effect by using conventional single input single output control system。In this case, the decoupling control  algorithm  for multivariable coupling control system emerges。

Firstly,In this paper, a diagonal-matrix is designed based on a dual-input and dual-output system and its simulation model is builded in Simulink。Then the paper does a lot of research on the PID neural network decoupling algorithm and designs a PID neural network decoupling algorithm for a 3 input 3 output system with a complex coupling condition。 Two methods are put forward to improve the performance of the control system At the end of this paper, a PID neural network decoupling control method with particle swarm optimization algorithm is proposed。

The simulation results shows that the diagonal matrix decoupling algorithm can achieve a good control effect when the control system is not too complicated。 In the PID neural network decoupling system, the output can track the expected value of the system, the error value of the objective function is close to zero after a period of time。However there is still a large error in the steady state and the adjustment time is not fast enough。adding the momentum term and the neuron coefficient can improve the dynamic and static performance of the system to a certain extent。 The PID neural network with particle swarm optimization algorithm is more satisfactory in the aspects of steady state error and adjusting time。

Key words:PID neural network; multivariable control system; particle swarm optimization algorithm; decoupling control

目录

第一章 绪论 1

1。1 选题背景与意义 1

1。2 神经网络研究的发展及现状 2