基于后向回归网络的自适应多步预测

本文针对一类非线性动态系统,提出了一种新的基于后向回归网络的自适应多步预测方法,并对基于神经网络的自适应预测机理进行了分析。预测器由两个同构的后向回归网络来实现,输入及预测长度由随机读写存贮器单元的取值来控制。计算机仿真结果表明,这种自适应预测方法对一类恒定参数的非线性系统是可行的,可有效地处理系统具有的大时延和随机干扰。

Adaptive Multistep Prediction Based on Backpropagation Networks

This paper develops a novel backpropagation networks based adaptive multistep prediction technique for a class of nonlinear dynamical systems, and the prediction mechanism is analyzed. Two isomorphic backpropagation nets are used together to implement the proposed predictor. One is called learning network (LN) while the other is called prediction network (PN). The prediction length and the inputs to the two nets are controlled by settings of some random access memory (RAM) units. Simulation results show that this neural nets based adaptive predictor is applicable for a class of nonlinear systems with constant parameters, which may involve a large time delay and stochastic disturbances, so that it provides a new way to deal with a class of nonlinear prediction problems.