一种可分非线性系统的自校正广义预测

对一类可分非线性系统,采用Hammerstein模型的基本框架,用神经网络对非线性部分建模,线性部分采用受控自回归积分滑动平均模型。对此模型的线性部分设计广义预测控制器,得出线性部分的控制量。根据此控制量,引入一逆神经网络,结合原来的神经网络模型,通过对逆神经网络权值的调整,使神经网络模型的输出为线性部分的控制量,同时得到逆神经网络的输出,即非线性系统的控制量。克服了Hammerstein模型中非线性部分的反函数存在性和惟一性的问题。仿真结果验证了该设计的有效性。

Generalized Predictive Control of A Separable Nonlinear System

To a kind of separable non-linear system, based on the basic frame of Hammerstein model,the non-linear part adopts the neural network modeling and the linear part adopts CARIMA. To the linear part of the model this paper designs a generalized predict controller and draws control value.According to this control value , only updating weight values of INN , make output of NN model approach control value of linear part, at the same time get the output value of INN , namely the control value of the non-linear system.This method overcomes the existence of inverse function of the non-linear part and question of the uniqueness in Hammerstein model. A simulated study has proved its validity.