一类基于神经网络的非线性模型预测控制

在研究非线性对象输入/输出数据的基础上，将对象输出的Taylor级数展开式取线性项作为预测模型，提出一种非线性系统模型预测控制算法，为了保证预测模型的准确性，以神经网络做辩识器估计系统建模误差，对非线性对象进行单频预测控制，理论上已证明三层BP网能任意逼近L2上的非线性函数，本文通过仿真研究也表明了当神经网络逼近系统建模误差时，所提出的预测控制算法对复杂非线性对象能达到良好的控制效果。

Predictive Control of Nonlinear Model Based on Neural Network

Based on studies of input and output data, a predictive control method of nonlinear model is presented with the linear part of Taylor series expansion of nonlinear system used as predictive model. In order to ensure the accuracy of the predictive model, neural network is used as identifier to estimate the error of predictive model in one-step predictive control. It has been proved in theory that three-layer BP network can approximate any nonlinear function in L2 space arbitrarily and simulation samples in this paper also show that well control performance can be earned in this proposed NLMPC algorithm when the neural network can approximate the error of predictive model.