Nonlinear system identification based on generalized adaptive exponential function link networks model

Abstract

Extension functions based on the model of the trigonometric functional link networks (TFLN) and the adaptive exponential functional link networks (AEFLN) have been widely applied in nonlinear identification systems. However, these models are lack of cross-terms (multiple input sample and its past samples). This degrades their performance, especially in nonlinear systems containing strong nonlinear distortion. In this paper, we propose a generalized AEFLN model (GAEFLN) for nonlinear system identification. Since the GAEFLN contains sine, exponential, and cross-terms expansion function, its convergence characteristics will be significantly improved. Simulation results based on nonlinear system identification show that the performance of the proposed GAEFLN is superior to those of the TFLN and AEFLN.