Current Prediction of SPMSM Drives by Using Augmented ControlSynth Neural ODEs

Predicting the current evolution of SPMSM is crucial for assessing the operational safety of the motor. Accurate current predictions can effectively prevent safety issues in realworld scenarios. This model can be viewed as a continuoustime dynamic system, and in general, ControlSynth Neural ODE (CSODE) is effective in learning such models. However, under certain conditions, CSODE may encounter difficulties when learning these models. This paper proposes an augmented ControlSynth Neural ODE (ACSODE) to learn the model of SPMSM and demonstrates that, under certain conditions, traditional CSODE fails to effectively learn the nonlinear model, while the ACSODE can learn it accurately. The practicality of the ACSODE is validated through experiments on an SPMSM model with load disturbances. The experimental results show that the proposed method effectively and accurately reconstructs the system, demonstrating strong short-term and long-term prediction capabilities. This study provides new insights for learning complex, data-driven dynamic models and has further application potential in fields such as robotics and electrical machines.