Vehicular Lead-Acid Battery Fault Prediction Method based on A-DeepFM

Vehicle lead-acid battery failures can severely impact automotive safety and normal usage. Early detection of potential faulty batteries has become an important issue in the industry. This paper addresses the complex operating conditions of vehicle lead-acid batteries, the large volume of battery data with few signal dimensions, low efficiency in fault point annotation, and inadequate prediction accuracy. A deep learning-based fault prediction method using multi-dimensional time series data from vehicle lead-acid batteries is proposed. By employing an automatic fault segment annotation method, manual feature design, and an improved A-DeepFM model, the performance of the battery fault prediction task is optimized. Finally, on an independent test set containing 10000 batteries, the results show that the A-DeepFM model achieves a prediction Precision of 93% in the vehicle lead-acid battery prediction task, improving by more than 2.2% compared to expert strategies and model methods and able to predict battery failures on average 5.3 days in advance, indicating a clear advantage of the proposed method.