Optimization of an ANN-based speed and position estimator for an FOC-controlled PMSM using genetic algorithm

Abstract

To further improve the performance of sensorless permanent magnet synchronous motor (PMSM) implementations, this study develops a neural network-based estimator for the speed and position estimation of a PMSM using field oriented control (FOC) as its control scheme. The proposed neural network’s hyperparameters are optimized using genetic algorithm. The neural network is trained and optimized based on a training dataset obtained from the Simulink simulation of the motor control system. The hyperparameters optimized include the training algorithm parameters, batch size, and the number of hidden layers and the corresponding neurons. The proposed estimator performed with better estimation accuracy than conventional estimators such as the sliding mode observer (SMO), model reference adaptive system (MRAS), and two other neural network configurations. The qualifications were made on steady-state and dynamic conditions. In terms of efficiency, the proposed estimator has a relatively lower power consumption but still falls short of the power drawn when using an actual sensor. The qualification process verified that the optimization of the neural network’s hyperparameters using genetic algorithm can provide a better performance in the estimation of motor parameters in sensorless motor applications.