Enhanced prediction for battery aging capacity using an efficient temporal convolutional network

For electric vehicles (EVs), reliable prediction of battery aging capacity is crucial to guarantee the driving safety. However, advanced prediction methods incur more energy consumption because of their substantial computation requirements. Consequently, a compact and accurate prediction model is urgently needed. To this end, we propose an efficient prediction model that simultaneously achieves the dual goals of reducing the computational burden and improving the accuracy of its predictions. From the extensive real-vehicle data, the charging hot-spot ranges are divided to establish a reliable foundation for screening high-quality and low-quantity data. Based on the screened data, a seasonal-trend decomposition algorithm and an independent feature analysis are presented to reveal significant features associated with aging capacity. Subsequently, we combine the advantage of the decomposition method and a temporal convolutional network (DMTCN) to predict the battery aging capacity with satisfactory accuracy in the near future. Furthermore, a particular decoupling mechanism is proposed to rectify the analytical interference and misinformation caused by the mixing of data. The experimental results demonstrate that the computational burden of the proposed DMTCN model is obviously reduced using this mechanism, while the prediction accuracy is further enhanced. We are delighted that this new attempt has achieved the impressive results.