Nonlinear Modeling of Switched Reluctance Motor Based on BP Neural Network

Abstract

Due to highly nonlinear characteristics of switched reluctance motor (SRM), an accurate nonlinear model is the key to minimize torque ripple by optimum phase current profiling. After static torque characteristics of SRM having been measured by DSP, the inverse model of torque is developed based on BP neural network The networks are trained with several improved algorithm. It is found that for the nonlinear model of SRM, the Levenberg-Marquardt (LM) algorithm has faster convergence and more accuracy than the other techniques on BP neural network Compared with experimental dado, accuracy of the inverse model of torque for SRM based on BP neural network with LM algorithm is proved With this model, the torque ripple minimization can be achieved by optimum profiling of the phase current based on instantaneous torque control. Simulation results verify the feasibility of this torque ripple minimization technique.