Smart Health State Diagnosis of Lithium-Ion Batteries Using Wavelet-Enhanced Hybrid Deep Learning with an Attention Mechanism

Abstract

The growing reliance on lithium-ion batteries in electric vehicles has driven significant research efforts aimed at improving the safety and performance of Battery Management Systems (BMS). Central to the effectiveness of a BMS is the precise determination of the State of Health (SOH), which plays a key role in guaranteeing the consistent functionality of electric vehicles. To tackle this issue, we introduce a time-series prediction model specifically designed for assessing the SOH of lithium-ion batteries. Our proposed AM-CNN-WBiLSTM framework integrates a Convolutional Neural Network (CNN) with a Wavelet-enhanced Bi-directional Long Short-Term Memory (WBiLSTM) component, further refined by an attention mechanism to boost forecasting accuracy. This approach achieved exceptional results, with a Mean Absolute Percentage Error (MAPE) of 0.25% and a Root Mean Square Error (RMSE) of 0.26%, surpassing existing methods. These outcomes mark a substantial leap forward in predicting battery health and efficiency for electric vehicles, supporting the wider uptake of sustainable and reliable electric mobility solutions.