Battery life prediction based on LSTM segmented iteration

Lithium-ion batteries are widely used in electronic devices and electric vehicles, mainly due to their high energy density and long cycle life. However, the performance of Li-ion batteries is susceptible to gradual degradation due to factors such as ambient temperature fluctuations, material aging, and the cumulative effects of charge/discharge cycles. Although the traditional long short-term memory network (LSTM) is capable of capturing long-range dependencies intime series, it is susceptible to random initialisation and has insufficient global search capability. In light of the aforementioned considerations, this thesis proposes a segmented iterative LSTM model based on Bayesian search. This approach preserves the time series features through a segmented iterative strategy, captures the changes in different decay stages of the battery, and optimises the hyperparameters through the use of Bayesian search, thereby enhancing the stability and prediction accuracy of the model. Experiments using the NASA lithium-ion battery dataset show that the model outperforms the traditional LSTM architecture in terms of root mean square error (RMSE) and mean absolute error (MAE). It demonstrates enhanced generalisation ability and prediction accuracy, offering a novel approach to battery life prediction.