Parameter Tuning for Comparing Multi-Objective Evolutionary Algorithms Applied to System Identification Problems

Abstract

The parameter estimation tuning is one of the procedures of system identification. One of the identification problems is to find an optimal model structure especially for representing the complex multivariable dynamic systems. Therefore, this problem needs to be solved by using multi-objective optimization with two objective functions namely minimum predictive error and model complexity. In this paper, the tuning of control parameters is studied before comparing multi-objective evolutionary algorithms. The framework is based on the system identification problems. In parameter tuning, two main qualitative parameters of evolutionary algorithms are focused on crossover rate and mutation rate. The multi-objective evolutionary algorithms used for this study are an elitist non-dominated sorting genetic algorithm (NSGA-II) and multi-objective optimization using differential evolution (MOODE). The different parameter settings are compared in order to obtain good parameter setting for NSGA-II and MOODE. Then, the performance for both algorithms is also compared using Pareto front. The plotted graphs of two objective functions show the non-dominated fronts achieved by the algorithms. The results proved that MOODE shows an advantage in solving system identification problems in a multivariable nonlinear dynamic system.