Ｈａｍｍｅｒｓｔｅｉｎ模型基于神经网络的预测控制方法

Ｈａｍｍｅｒｓｔｅｉｎ模型是化工过程中最常用的模型之一，它由非线性静态环节和线性动态环节串连 组成，适合描述ｐＨ过程和具有幂函数、死区、开关等非线性特性的过程．这类模型的控制问题可以分解 为：线性模型的控制问题和非线性模型的求根问题．针对Ｈａｍｍｅｒｓｔｅｉｎ模型提出了一种基于神经网络的 模型预测控制策略，采用一组神经网络拟合非线性部分的逆映射．这种方法不需要假设Ｈａｍｍｅｒｓｔｅｉｎ模 型的非线性部分由多项式构成，并且避免已有研究在无根和重根情况下存在的问题．最后通过仿真试验证明了以上结论．

Model predictive control based on neural networks for Hammerstein type nonlinear systems

The Hammerstein model is composed of a nonlinear static element and a linear dynamic element serially,and it proves to be effective in describing the behavior of many chemical processes.By appropriate identification,the intricate nonlinear control problem of this model can be facilitated into two problems:the control of the linear part and the solution of the nonlinear part.In this paper,a model predictive control scheme is proposed,which uses a set of neural networks to approximate the inverse mapping of the nonlinear block.This neural networks method needn't assume that the nonlinear block is a polynomial equation,thus it overcomes the difficulty that no real roots exist for the polynomial equation.Two simulation examples,including a pH neutralization process,are used to demonstrate the effectiveness of the method.