Position sensorless control of switched reluctance motors based on improved neural network

Abstract

Neural network (NN) has been applied as rotor position estimator in switched reluctance motor (SRM) whose characteristic is highly non-linear. However, conventional two inputs back propagate (BP) NN based rotor position estimator is not appropriate for real-time application in high speed operations because of its considerable computational time consumption in its hidden layer. In this study, an improved BP NN with inductance input pretreatment for the rotor position estimator of SRM is proposed. It achieves 75.44% computational burden reduction while staying at the same accuracy as the conventional one. Moreover, with the pretreatment, the NN can output the angle of full electrical period of the SRM operation. Training data includes rotor position, flux-linage which is acquired by finite element analysis (FEA) and phase current Sensorless control algorithm is also described. Simulations and experiments are performed based on a 12/8 SRM. The results are compared with conventional method. The effectiveness of the proposed sensorless estimator and control strategy are testified under low speed, high speed and sudden loading change operations.