基于神经网络PI控制器的设计.docVIP

基于神经网络的PID控制器设计 摘 要 本文以提高控制器的控制效果为目标，将神经网络与PID控制相结合，分别对单变量系统和多变量系统的神经网络结构的PID控制进行了深入研究和探索。 对单变量系统，将预测控制的思想和神经网络PID控制的思想结合起来，用多步预测性能指标函数去训练神经网络把纬d器的权值，就构成了多步预测性能指标函数下的神经网络PID控制系统:本文对此系统进行了改进，引入了新的多步预测性能指标函数，同时对神经网络辨识器部分，采用了更适合于实时控制的动态递归神经网络来代替原方案中的多层前向网络对其辨识。仿真结果表明此改进方法比原方法及带辨识器的神经网络PID控制方法具有更好的响应性能。 对多变量系统，首先研究了基于多步预测性能指标函数下多变量系统的神经网络PID控制，并给出仿真实例及结论;接着研究了PID神经元网络多变量控制器的结构和计算方法:它是由并列的多个子网络组成，当控制系统有n个被控变量，子网络就有n个。每个子网络的输入层接受系统的给定信号和对象的输出信号;隐含层由比例元、积分元、微分元组成，实现PID运算;输出层实现规律的综合:为加快权值调整速度对输出层权值采用最小二乘法进行调整，代替原方案中的梯度法，仿真结果表明此系统具有良好的自学习和自适应解祸性能。 关键词: 神经元网络 PID控制 多步预测性能指标函数 动态递归神经网络 单变量系统 多变量系统 The PID Controller Was Based On Neural Network Design ABSTRACT In order to enhance the performance of the controller, this paper combined the neural network and PID control, and deeply studied the neural network Pm controller based on single-variable and mufti-variable system. For single-variable system, the neural network PID controller based mufti-step predictive performance target function combined the predictive control idea and the neural network PID control idea. It uses multi-step predictive performance target function to train the weights of neural network PID controller, This paper improved the system: It uses new mufti-step predictive performance target function to train the weights, and it uses the dynamic recursion neural network instead of multiplayer feed forward neural network that is furthermore fit for real-time control to identify the part of neural network identification. The simulating results shows that this method has better response performance than the neural network PID control method with identification For mufti-variable systems, at first, This paper studied the neural networks PID controller based mufti-variable systems using mufti-step predictive performance target function, After studying the systems simulating instances, I got the results; Then this paper studied the structure and arithmetic of the PID neural network multivariable controller. I