Data-Driven Based Hybrid Predictive Model for the PMSM Drive System

Permanent magnet synchronous motor (PMSM) drive system is a time-varying nonlinear system that integrates several physical domains, including mechanical, electrical, and electromagnetic. Thus, obtaining an accurate mathematical model of PMSM drive system is an easily overlooked challenge. In this paper, a data-driven based machine learning approach is introduced to model the dynamics of PMSM drive system. Compared to traditional mathematical PMSM model, it does not include initial parameters and any assumptions. In this paper, a time series datasets of the drive system are constructed for the whole operating range of the PMSM. And then, the Pearson correlation is adopted to investigate the coupling between variables of PMSM states. To predict the PMSM states, a hybrid predictive models based on the long-short term memory and transformer are proposed. The data of dq-axis currents, speed and electromagnet torque can be obtained by feeding the data of voltage variables into the hybrid predictive models. Finally, the test results show that the proposed hybrid predictive model can accurately predict the temporal dynamics of PMSM drive system in real time.