Lithium-ion battery remaining useful life prediction based on interpretable deep learning and network parameter optimization

As intelligent computation power in embedded systems has rapidly developed in recent years, the health state monitoring and remaining useful life prediction of batteries based on deep learning can gradually be deployed and applied in the onboard management system. However, there are still problems with large amounts of data calculation, high model complexity, and poor interpretability. Therefore, this paper proposes a remaining life prediction method for batteries combined with interpretable deep learning and network optimization. First, based on the fused deep learning model, the interpretable algorithm is used to explain the degree of attention of the model to different features and quantify the contribution of each part in input data, thereby identifying important aging features and removing useless data. Then, structured pruning is adopted to remove redundant network parameters under the constraints of ensuring prediction accuracy. The structure generally realizes model interpretation and full process optimization from battery aging data to network parameters. According to the validation of the selected dataset, compared with the original model, the model optimized by the method proposed in this paper has an average prediction accuracy increase of 0.19 % and an average speed increase of 46.88 %. It greatly saves computational resource consumption and improves model operation efficiency while ensuring prediction accuracy. In addition, the explanation and analysis of crucial feature areas in battery aging data provide a reference for effective health management.