Premium
A multiple‐input–single‐output fractional‐order Hammerstein model identification based on modified neural network
Author(s) -
Jahani Moghaddam Mohammad,
Mojallali Hamed,
Teshnehlab Mohammad
Publication year - 2018
Publication title -
mathematical methods in the applied sciences
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.719
H-Index - 65
eISSN - 1099-1476
pISSN - 0170-4214
DOI - 10.1002/mma.5136
Subject(s) - weighting , nonlinear system , convergence (economics) , mathematics , artificial neural network , radial basis function , transfer function , function (biology) , control theory (sociology) , system identification , radial basis function network , identification (biology) , algorithm , nonlinear system identification , mathematical optimization , computer science , artificial intelligence , data modeling , biology , control (management) , botany , economics , database , economic growth , engineering , quantum mechanics , evolutionary biology , radiology , medicine , physics , electrical engineering
This paper presents a new multiple‐input–single‐output nonlinear system identification method based on Hammerstein model, which includes a fractional transfer function and a Modified Radial Basis Function Neural Network (MRBFNN) as linear dynamic part and static nonlinear subsystem, respectively. The size of Radial Basis Function Neural Network (RBFNN) grows with the number of inputs exponentially. As a novel idea, the MRBFNN is proposed, whose adjustable parameters are far fewer than other RBFNNs presented yet. A Modified Genetic Algorithm is used to identify the fractional orders and the centers and widths of MRBFNN and obtain an initial estimation of other unknown parameters. The Recursive Least Square (RLS) method is used to improve the estimation by updating the weighting parameters of MRBFNN and the transfer function coefficients. The convergence analysis of the proposed RLS is provided. Simulation results show the effectiveness and accuracy of the proposed method.