Data-Driven Modeling and Compensation Strategy of PMSM Considering Core Loss and Saturation

Accurate model of permanent magnet synchronous machine (PMSM) is significant for high-performance control. The machine model can be affected by various factors such as magnetic saturation and core loss effect, especially in the deep saturation and high-speed regions. This article proposes a data-driven-based machine modeling and compensation approach to improve the model accuracy by considering saturation and core loss effect. In the proposed approach, magnetic saturation is initially modeled using nonlinear polynomials and core loss effect is modeled with various speed data. The model mismatch due to these effects is then derived to generate the training data for the neural network (NN), which can accurately predict the model mismatch under various operating conditions. In comparison to the conventional model, the proposed approach adds compensation terms directly to the machine models, which can achieve better accuracy with efficiency and simple implementation, which can be utilized in motor control and parameter estimation. The proposed approach is validated on a laboratory interior PMSM and compared with existing methods under various operating conditions.