Design of machine learning-based controllers for speed control of PMSM drive

Abstract This study presents machine learning (ML)-based controllers for a surface permanent magnet synchronous motor (PMSM) drive system. The ML-based regression techniques like linear regression (LR), support vector machine regression (SVM), feedforward neural network (NN) and advanced NN like Long Short-Term Memory network (LSTM) are explored here in detail. This paper aims to develop an improved vector controller based on machine learning, and to investigate ML algorithms which are not yet been explored for the current control of a PMSM drive. The proposed machine learning-based control approach, which explores the influence of decoupling terms on vector control, is theoretically investigated and simulated in the vector control environment of the PMSM drive. The performance is also evaluated in real-time using the Opal-RT setup. The proposed control approach demonstrates the ability to fulfill the speed tracking requirements in the closed-loop drive system. A comparison of the simulation results between the PI controller and the suggested control algorithms validates the effectiveness of the proposed control algorithms for speed control applications. The performances of the proposed ML-based controllers improved in terms of evaluation metrics, transient peak levels and current responses, when compared to the conventional PI controller.