A Fault Prognosis Model of Battery Packs of Electric Vehicles based on Long Short-term Memory Neural Networks

As electric vehicles (EVs) begin to gain popularity, the damage caused by battery packs of EV s has raised concerns. This paper proposes a fault prognosis model based on long short-term memory (LSTM) neural networks to ensure the safety of EVs. The proposed model reserves time for fault handling by predicting whether the fault will occur in the battery pack of EV after a period of time. Battery data obtained from the battery management system (BMS) is used for model training and verification. Meanwhile, a data augment method called “stratified sampling” for obtaining more battery data is proposed in this paper. Traditional prognosis models require many feature engineering works based on electrochemical expertise and ignore the time dependencies hidden in battery time-series data. The proposed model builds an end-to-end structure from the original data directly to the prognosis result to address these issues. The model consists of two parts: LSTM network is used for capturing long-term dependencies in the data; a convolution operation performs higher-level feature extraction on the data. Experimental results show that the proposed fault prognosis model implements a general method to predict different types of battery faults, with an accuracy rate of over 82.7%.