An improved model of switched reluctance motors based on least square support vector machine

Abstract

The least square support vector machine (LSSVM) is a powerful statistic learning technique used to solve various nonlinear problems and has gained attentions in modeling of SRMs. But the conventional SRM model based on LSSVM has two defects, one is the large error of the torque owing to the cascade of single-output LSSVMs, the other is the low generalization ability of the SRM model caused by the selected Radial Basis Function (RBF) Kernel function. An improved SRM model based on LSSVM constructed by improving the structure and algorithm is proposed. It has been proved that the improved SRM model has overcome the shortages of the conventional LSSVM models and got better learning and generalization abilities and less time consuming. Results of simulation and experiment show good consistency and real-time, demonstrating the effectiveness and high accuracy of the proposed model.