Parameter Optimization for Model-Free Predictive Current Control of PMSMs Using ANN and K-Fold Cross-Validation

Model-free predictive current control (MFPCC) of permanent magnet synchronous motors (PMSMs) based on an extended state observer (ESO) has been extensively studied in recent years due to its simplicity, parameter efficiency, and strong system robustness. However, the performance of MFPCC-ESO is significantly affected by the selection of internal parameters. To address this challenge, this paper proposes an artificial neural network (ANN) approach combined with K-fold cross-validation for parameter optimization in MFPCC-ESO. Firstly, extensive simulations are conducted using different parameter selection schemes, and key performance metrics, including speed recovery time and Total Harmonic Distortion (THD) of the current, are extracted. Next, a suitable ANN structure is designed based on K-fold cross-validation, and the collected data is used to train the network. In the final step, a fitness function is defined, and by leveraging the predictive capabilities of the ANN, refined parameter combinations are evaluated to identify a globally optimal parameter set that satisfies the fitness function. Through simulations, it is demonstrated that the proposed method outperforms conventional parameter selection schemes, achieving improved performance.