Capacity degradation prediction of on-road vehicle battery packs by combining Kolmogorov–Arnold with squeeze-and-excitation networks

Abstract With the rapid development of the electric vehicle (EV) market, accurate prediction of battery capacity degradation is crucial for ensuring the safe and reliable operation of EVs. Deep neural networks can automatically extract intrinsic features from data and perform well in handling complex nonlinear and high-dimensional data, providing a new direction for predicting battery capacity degradation. Kolmogorov–Arnold networks (KANs) based on the Kolmogorov–Arnold representation theorem have attracted attention because of their novel architecture and advantages in accuracy and interpretability. Combining KANs with squeeze-and-excitation (SE), a model termed SE-KAN is proposed to predict the capacity degradation of batteries. In this model, the SE module is used to enhance the network learning ability through adaptive feature recalibration, whereas the KAN module is used to improve the model’s adaptability to complex dynamic changes in charging data. This model is tested on a dataset consisting of real-world charging data from 20 vehicles. Thirty-three features are selected via the gray relational analysis method. Tests in which 50% of the samples are used for training reveal that the root mean square error and the mean absolute percentage error are 0.417 Ah and 0.167%, respectively. The comparison results show that this model outperforms the KAN and other comparative models.