Performance Comparison of Variants of Hybrid FLANN-DE for Intelligent Nonlinear Dynamic System Identification

Abstract

This work presents the performance comparison of different variants of hybrid Functional Linked Artificial Neural Network (FLANN) structures and Differential Evolution (DE) algorithm (FLANN-DE) for intelligent nonlinear dynamic system identification. FLANN is single-layer artificial neural network structure having less computational complexity and preferred for online applications and DE being a derivative-free metaheuristic algorithm is used as a global optimization tool. System identification finds its application in direct modelling, channel identification and estimation, geological exploration, instrumentation and control. Direct modelling is based on adaptive filtering concept and can be developed as an optimization problem. The goal of direct modelling is to estimate a model and a set of system parameters by minimizing the prediction error between the actual system output and the model output. The identification problem involves the construction of an estimated model which generates the output which matches that of desired system output when subjected to the same input signal. In this present work, hybrid FLANN-DE is proposed for direct modelling of nonlinear dynamic systems and comparative analysis is carried out for different variants of FLANN structures such as Chebyshev FLANN (CFLANN), Legendre FLANN (LFLANN) and Trigonometric FLANN (TFLANN) in terms of performance, the speed of computation and accuracy of results.