Self-organizing interval type-2 fuzzy neural network based on eigenvalue decomposition

In this study, an eigenvalue decomposition-based self-organizing interval type-2 fuzzy neural network (ED-SOIT2FNN) is proposed to tackle the identification problem of nonlinear systems. The network model determines both the structure and parameters of the network through online learning, which can...

Full description

Saved in:

Bibliographic Details
Published in:	Applied soft computing Vol. 184; p. 113741
Main Authors:	Wang, Panchao, Zhao, Taoyan, Cao, Jiangtao, Li, Ping
Format:	Journal Article
Language:	English
Published:	Elsevier B.V 01.12.2025
Subjects:	Adaptive discount recursive partial least square algorithm Eigenvalue decomposition (ED) Self-organizing interval type-2 fuzzy neural networks Sliding window adaptive second-order algorithm with forgetting factor ε-completeness criterion Sliding window adaptive second-order algorithm with forgetting factor Adaptive discount recursive partial least square algorithm Self-organizing interval type-2 fuzzy neural networks ε-completeness criterion Eigenvalue decomposition (ED)
ISSN:	1568-4946
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In this study, an eigenvalue decomposition-based self-organizing interval type-2 fuzzy neural network (ED-SOIT2FNN) is proposed to tackle the identification problem of nonlinear systems. The network model determines both the structure and parameters of the network through online learning, which can realize the structure learning and parameter learning simultaneously. Firstly, in the structure learning process of ED-SOIT2FNN, the error criterion and the completeness criterion of fuzzy rules are used to verify whether the rules grow. Meanwhile, the eigenvalue decomposition method is adopted to find the less active rules for deletion, so that obtain a more compact network structure. Secondly, in terms of ED-SOIT2FNN parameter optimization and the characteristics of network parameters, they are divided into the linear and nonlinear ones. The algorithm of adaptive discount recursive partial least square is employed to optimize the linear parameters, which is conducive to improving the noise resistance of the network model and solving the data saturation problem. And the sliding window adaptive second-order algorithm with a forgetting factor is adopted to optimize the nonlinear parameters. Compared with the algorithm of gradient descent optimization, it can accelerate the convergence and achieve a good adaptability with stability. Finally, the proposed ED-SOIT2FNN was applied to four typical nonlinear examples for identification. The experimental results showed that compared with similar methods in the existing literature, the proposed ED-SOIT2FNN could produce a more compact network structure with higher accuracies of identification and prediction. •The error criteria and the completeness criteria of fuzzy rules were used to determine.•The algorithm of adaptive discount recursive least square was integrated to optimize the network parameters of ED-SOIT2FNN.•The proposed ED-SOIT2FNN was applied to typical examples of nonlinear system identification.
ISSN:	1568-4946
DOI:	10.1016/j.asoc.2025.113741