NN-based model predictive direct speed control of PMSM drive systems

Abstract

PMSM servo systems require a high dynamic on speed control. In this paper, a modified model predictive direct speed control based on neural network is proposed, which not only overcomes limitations of cascaded linear controller, but also improves the poor adaptability of model predictive direct speed control(MP-DSC) based on mathematic model. BP neural network (NN) approaches the dynamics of PMSM, and predicts the future speed. The finite control set approach is applied to select input switch states of inverter directly depending on the predicted speed error. Moreover, The BP neural network is trained through the online sliding-window learning, which can make the BP neural network estimates the local dynamics of the system using limited input-output datum in the window, and is more suitable for online realization. Simulation and experiment have verified that the proposed control strategy can not only achieve promising dynamic and steady behavior, but also excellent adaptability to parameter perturbation and external disturbance.