一种用于非线性动态辨识的新型神经网络  

A New Nonlinear Dynamical Identification Neural Network

作　　者：张剑 林瑞昌 毕天昊 ZHANG Jian;LIN Ruichang;BI Tianhao(School of Intelligent Manufacturing,Guangzhou Panyu Polytechnic,Guangzhou 511483,China;College of Engineering,National Yunlin University of Science and Technology,Yunlin 640301,China)

机构地区：[1]广州番禺职业技术学院智能制造学院,广东广州511483 [2]台湾云林科技大学工程学院,中国台湾云林640301

出　　处：《控制工程》2024年第8期1383-1391,共9页Control Engineering of China

摘　　要：为提高非线性动态系统辨识(NDSI)的效果,在结合自建型模糊神经网络(SCFNN)和多层神经元神经网络(MLPNN)的基础上,提出一种自建递归型模糊神经网络(SCRFNN)。SCRFNN相较于前者,多了一个递归通道与抑制模糊规则产生机制;相较于后者,增加了模糊推论与一个递归通道。为验证SCRFNN在系统辨识中的有效性,设计一个新的NDSI在线学习模型与代码设计流程图,并以此作为在线学习架构,将以上3个神经网络模型对4个串-并型非线性动态系统进行辨识分析。经过仿真表明,新提出的SCRFNN通过存储内部状态,具备了映射动态特征的功能,从而使系统具有适应时变特性的能力,更适合于非线性动态系统的辩识。且在模糊规则数、学习收敛速度、学习与预测误差均方根值、预测精准度方面也取得了良好的效果。In order to improve the results of nonlinear dynamic system identification(NDSI),on the basis of combining self-constructing fuzzy neural network(SCFNN)and multi-layer perceptron neural network(MLPNN),a self-constructing recurrent fuzzy neural network(SCRFNN)is proposed.Compared with SCFNN,SCRFNN has one more recursive path and a mechanism to suppress the generation of fuzzy rules.Compared with MLPNN,SCRFNN adds fuzzy inference and a recursive path.To verify the effectiveness of SCRFNN,a new NDSI on-line learning model and code design flow chart are designed.Based on the model,four series-parallel nonlinear dynamic systems are identified by SCRFNN,SCFNN and MLPNN.Compared with the static feed forward neural network,the simulation results show that the SCRFNN has the function of mapping the dynamic characteristics by storing the internal state,so that the system has the ability to adapt to the time-varying characteristics,and is more suitable for the identification of nonlinear dynamic system.To face the same NDSI,SCRFNN is also better than SCFNN and MLPNN in fuzzy rules,learning convergence speed,learning and prediction error root mean square value,and prediction accuracy.

关 键 词：自建递归型模糊神经网络 自建型模糊神经网络 多层神经元神经网络 非线性动态系统辨识 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]