基于动态递归模糊神经网络的动态系统辨识

来源期刊：中南大学学报(自然科学版)2003年第3期

论文作者：张友旺

文章页码：277 - 280

关键词：动态递归;模糊神经网络;动态反向传播学习算法;动态系统;辨识

Key words：dynamic recurrence; fuzzy neural network; dynamic back propagation algorithm; dynamic systems;identification

摘    要：模糊系统和神经网络由于具有逼近任意连续非线性映射的特性而广泛应用于系统的辨识和控制,但是传统的模糊神经网络是一种静态映射,不适用于动态系统的辨识,而现实工程中的控制对象反映的是系统的动态行为.为了提高动态系统的辨识精度,提出了一种新型的动态递归模糊神经网络,并根据动态递归神经网络的数学模型推导其动态反向传播学习算法及其改进算法.仿真结果表明:由于动态模糊神经网络的辨识过程同时利用了系统的当前数据和历史数据,对动态系统的辨识,特别是对具有纯时间延迟动态系统的辨识,较传统模糊神经网络在辨识精度和稳定性方面具有更好的效果.同时,确定网络权值和隶属函数参数初始值的方法可使动态系统的辨识过程具有更快的收敛速度.

Abstract: The fuzzy system and neural network have the property of approximating any continuous functions and are widely used for system identification and control, but the traditional fuzzy neural network is a kind of static map and not suitable to identify the dynamic system, and the controlled plant in the practical engineering reveals the dynamic behavior. In order to improve the identification accuracy of the dynamic system, a novel dynamic recurrent fuzzy neural network is presented, and its dynamic back propagation algorithm and modified algorithm are formulated based on its mathematical models. The simulation results show that the utilization of the current and past data of the system at the same time makes the presented dynamic recurrent neural network more effective than the traditional fuzzy neural network in view of accuracy and stability for the identification of the dynamic systems, especially for the dynamic systems with pure time delay. The method of defining the initial value of the weight and the parameters of member functions makes the identification process converge faster.