Application of LSSVM with AGA Optimizing Parameters to Nonlinear Modeling of SRM

Abstract

Considering nonlinear magnetization characteristics of a switched reluctance motor (SRM), this paper presents a nonlinear model of SRM based on the integration of least square support vector machine (LSSVM) and adaptive genetic algorithm (AGA), known as LSSVM-AGA. The real-valued AGA is applied to optimize the parameters of LSSVM, and then the LSSVM using the optimal parameters forms a very efficient mapping structure for the nonlinear SRM. The hybrid method for modeling SRM is tested through sufficient sample data to verify its validation and feasibility. The sample data comprise flux linkage, current and rotor position, which obtained from the experimental SRM by the dc-excitation method. The forecasted data of the SRM model with LSSVM-AGA are compared with measured data, and error analyses are given to determine performances of the model. The experimental results demonstrate that LSSVM optimized by AGA performs better forecast accuracy and successful modeling of SRM.