一种基于神经网络辨识的预测方法

针对时变的非线性系统,将传统的预测控制与神经网络逼近任意非线性函数的能力相结合,提出一种基于神经网络辨识的预测方法.同时选用含有调整参数的双曲正切函数作为节点的激活函数,弥补由于未考虑激活函数的输出值域、影响神经网络辨识精度和速度不足的问题.仿真结果表明,它适用于无滞后和有滞后的时变非线性系统,辨识收敛速度快、精度高,并具有较强的自适应性和鲁棒性.

A Predictive Method Based on BP Neural Network Identification

Aiming at the time-varying nonlinear system,we propose a predictive method based on BP neural network identification,which combines the traditional predictive control with the neural network's capability of approximating to the non-linear function.Meanwhile,the hyperbolic function with adjustable parameters is selected as the activation function of nodes.This compensates the deficiency of Neural Network identification in accuracy and speed,caused by neglecting the output domain of the activation function.Simulation results show that the proposed method has an adaptive performance to the nonlinear system with or without delay,and it has a high speed in identification converges,with a higher accuracy,and a stronger adaptability and robustness.