KS-Autoformer: An Autoformer-Based SOC Prediction Framework for Electric Vehicles

Accurate state of charge prediction is essential for battery management systems, which is crucial for improving battery utilization efficiency and ensuring safety performance. Current SOC prediction models only consider battery-related features but ignores vehicle information. Additionally, there are challenges preventing gradient disappearance and explosion during model training due to excessive data and noise. This paper introduces a new framework that integrates laboratory battery data with vehicle features to improve the accuracy of SOC. First, we apply Matlab/Simulink to simulate an electric vehicle and process the generated vehicle data with spearman correlation analysis to identify the most relevant features, such as the electric motor, differential, and aerodynamic drag. Then we develop a data fusion model to synchronize the heterogeneous datasets with different frequencies to capture the sudden change in electric vehicles. Furthermore, we propose a KS-Autoformer prediction model to address the overfitting problem caused by data redundancy and noise, which improves accuracy by smoothing sensor noise and enhancing time delay aggregation in an auto-correlation mechanism. Finally, we utilize different driving cycles for training and testing in order to effectively evaluate the extrapolation and adaptability of our model. Experimental results show that our model achieves a significant improvement in predicting SOC in terms of accuracy and robustness.