Analysis of Excitation Current in Synchronous Motors Based on Multiple Machine Learning Models

This study focuses on accurately predicting the excitation current in synchronous motors using a hybrid machine learning algorithm. Given the critical role of excitation current in maintaining motor stability and performance, reliable predictive models are essential for optimal motor control and energy efficiency. This paper compares the predictive capabilities of Multi-Layer Perceptron (MLP), Random Forest (RF), and Support Vector Machine (SVM), MLP-RF and MLP-SVM. Experimental results demonstrate that the combined models outperform individual algorithms in both accuracy and robustness. The MLP-RF model, in particular, achieves the highest prediction accuracy on the validation and test sets, benefiting from MLP's adaptability and RF's ensemble averaging. Further analysis of the models' R^2 values across multiple trials reveals that the hybrid models offer more consistent predictions, making them less sensitive to variations in the data.