RBF and MLP Neural Network Speed Observer for Sensorless DTC Drive of IPMSM

RBF and MLP Neural Network Speed Observer for Sensorless DTC Drive of IPMSM

In this paper neural network speed observers for sensorless DTC drive of IPMSM are presented and comparisons between MLP and RBF neural networks inthis case, have done. Introduced neural network based speed observers are trained by Imperialist Competitive Algorithm (ICA). Due to artificial neural networkcharacteristics the proposed speed observers work in wide range speed as opposed to previous observers that doesn’t works in low speed or high speeds. Since neural network is trained with ICA, optimum weights of neural network are obtained. Simulation results on different conditions showthe good performance of proposed speed observers. However simulation shows that, RBFNN base speed observer has better performance than MLP neural networkobserver, both observer have good performance in wide range speed. In the other word operation in both low andhigh speeds is the main advantage of presented speed observers.