Nonlinear Modeling of a Switched Reluctance Motor using LSSVM - ABC

Abstract

To enhance the performance of a Switched Reluctance Motor (SRM), it is essential to build a precise model of the machine. Due to its double salient structure, the model is nonlinear. It is complicated to develop a simple mathematical model for such nonlinear systems. However, owing to the advancements in kernel based learning methods for statistical learning and structural risk minimization,the use of Least Square Support Vector Machine (LSSVM) for function estimation is powerful and accurate, with less computations needed. The accuracy of the regression model and generalization ability depends upon the proper selection of hyperparameters of LSSVM. It is necessary to employ a meta-heuristic technique for the optimal values of hyperparameters. This paper presents a nonlinear flux modeling of SRM based on LSSVM regression optimized by Artificial Bee Colony (ABC) algorithm. The training and testing sample data are obtained by 2D FEA using MagNET software. It was analyzed that the tuning by ABC algorithms offers a high degree of accuracy when compared to the values obtained by Particle Swarm Optimization (PSO), Differential Evolution (DE) and Genetic Algorithm (GA).