State of health prediction of lithium-ion batteries based on the HHO-BiLSTM-Attention model

Aiming at the problems in State of Health (SOH) prediction of lithium-ion batteries, such as insufficient extraction of indirect health features, insufficient utilization of temporal degradation information, and model hyperparameters relying on manual tuning, a prediction method based on the Harris Hawks Optimization (HHO) algorithm, the Bidirectional Long Short-Term Memory Network (BiLSTM) and the attention mechanism is proposed. First, multi-dimensional health factors are extracted from the constant-current constant-voltage charge-discharge curves and preprocessed; second, an HHO-BiLSTM-Attention prediction model is constructed, which uses BiLSTM to extract the temporal dependence features of health factors bidirectionally, adaptively weights the hidden states of each time step through the attention layer, and uses the HHO algorithm to globally optimize the key hyperparameters of the model. The experimental results on the public NASA dataset show that, compared with the BiLSTM and BiLSTM-Attention models, the error of the proposed model is significantly reduced, the mean absolute percentage error is below 0.6%, and the coefficient of determination reaches more than 0.95, demonstrating higher prediction accuracy and generalization stability. The model can accurately adapt to SOH prediction tasks of batteries under different operating conditions, and can provide effective technical support for the health assessment and safe operation and maintenance of power batteries.