Prediction of remaining driving range for electric vehicles based on IVY feature selection and parameter optimization of KAN

Abstract

Accurate prediction of the remaining driving range (RDR) of electric vehicles (EVs) is essential for enhancing the driving experience and optimizing battery management. However, existing studies still exhibit limitations in feature selection and model parameter optimization—two core steps in predictive modeling—thereby failing to maximize the predictive potential of key features. To address these challenges, this paper proposed an IVY Feature Selection and Parameter Optimization of Kolmogorov-Arnold Network (IVY-FSPOKAN) for predicting the RDR of EVs. The framework leverages the IVY algorithm to optimize feature combinations and the Kolmogorov-Arnold Network (KAN) structure and hyperparameters, thereby enhancing prediction accuracy and stability. The effectiveness of the IVY-FSPOKAN prediction framework is validated through experimental testing. The study first examined the impact of feature selection and parameter optimization on the prediction accuracy of RDR. The results show that both feature selection and parameter optimization significantly improve the accuracy of electric vehicle RDR predictions. Subsequently, the IVY-FSPOKAN framework was compared with 13 commonly used machine learning algorithms. The experimental results indicate that IVY-FSPOKAN outperforms the other algorithms across three evaluation metrics, with statistically significant differences. Finally, the Shapley Additive Explanations (SHAP) method was applied to analyze the importance of the features selected by IVY-FSPOKAN, intuitively revealing the mechanisms through which these features influence the RDR and providing interpretive support for the model's performance.