EvoNAS4Battery: An Evolutionary NAS Approach for Predicting the Remaining Useful Life of Lithium-Ion Batteries

Long-term use of lithium batteries inevitably leads to performance decay due to complex internal reactions and external interference, which can impact impacting battery lifespan and potentially causing equipment failure. Therefore, accurately predicting the remaining useful life (RUL) of batteries is crucial for predictive maintenance. While existing prediction methods based on deep learning have shown excellent performance, manually designing neural network structures remains a time-consuming and challenging task. To address this issue, we propose a neural architecture search (NAS)-based framework for battery RUL prediction. We introduce a novel network model based on the Transformer architecture to handle battery capacity regeneration interference and enhance time series information extraction. To efficiently find the optimal Transformer architecture, we use a NAS method assisted by a surrogate model as a predictor. Compared with the current state of research, extensive experimental results validate that our proposed method achieves the best overall performance.