Data-Driven Prediction and Feature Impact Analysis of Power Converter Performance in Electric Vehicles

Power converters, the backbone of electric vehicles (EVs), guarantee energy transfer and system reliability. Optimal design and operation depend on accurate prediction performance. An optimal Machine Learning (ML) regression model based on eXtreme Gradient Boosting (XGB) for forecasting power converter performance was developed. For closer prediction accuracy, three nature-inspired optimization algorithms, Crocodile Hunting Strategy (CHS), Chimp Optimization Algorithm (COA), and Weevil Damage Optimization (WDO), were combined. These algorithms iteratively tuned the hyperparameters of XGB, leading to better convergence and lower prediction errors. A thorough feature selection procedure identified the most significant parameters that induced model performance. The most influential factor affecting model performance was switching frequency (f-sw), with a feature importance score of 102.607, highlighting its decisive influence on converter efficiency. Efficiency ranked as the second most influential factor, with a score of 87.633, demonstrating its role in energy efficiency. Iteration count (iter) likewise exhibited immense influence with a score of 55.570 and highlighted computational iterations’ role in model stability. The integration of XGB and CHS, COA, and WDO resulted in superior prediction accuracy compared to conventional approaches. The optimized model advances understanding of decisive factors influencing converters’ performance and enables more efficient and more reliable power electronics for EVs. These results emphasize the hybrid ML and nature-inspired optimizations’capability to push EV technology further. Future work must examine more features and real-world testing to support further the model’s viability in various EV systems towards a wider implementation of sustainable transportation technologies.