Comparison of learning algorithms for neural network based speed estimator in sensorless induction motor drives

This paper identifies the suitable learning algorithm for neural network based on-line speed estimator in sensorless induction motor drives. The performance of sensorless controlled induction motor drives depends on the accuracy of the estimated speed. Conventional estimation techniques being mathematically complex require more execution time resulting in poor dynamic response. The nonlinear mapping capability and powerful learning algorithms of neural network provides an alternative for on-line speed estimation. A structurally compact and simple neural model is required for real time implementation to derive the desired accuracy and response time. This in turn to a large extent depends on the type of learning algorithm used to train neural based speed estimator. A self organizing Single Neuron Cascaded Neural Network (SNC-NN) architecture is trained off-line using three types of learning algorithm namely Backpropagation with Momentum (BPM), Variable Learning Rate (VLR) and Levenberg-Marquardt (LM) Algorithm to efficiently model the on-line speed estimator. The performance of the proposed NN based speed estimator model trained off-line with three different learning algorithms is compared in terms of accuracy, epochs needed for training and structural compactness. The suitable learning algorithm for off-line training of on-line NN based speed estimation in sensorless induction motor drives is identified and the promising results obtained are presented.