The rapid development of the electric vehicle industry has prompted much attention to the assessment of power battery health status. In this study, an intelligent prediction model of battery performance decline trend is proposed based on real vehicle data. Firstly, the capacity increment analysis is used to extract health features from the battery charging process, which is processed by a double filter algorithm to obtain a smooth capacity increment curve, and ten key health features are extracted. Subsequently, the correlation between the features and the battery health state is evaluated by Pearson correlation analysis, and the study shows that the correlation coefficient of most of the health features is greater than 0.85, which verifies the effectiveness of the feature extraction. Based on this, a GRA-EMD-BILSTM prediction model incorporating the attention mechanism was constructed, which utilized empirical mode decomposition to decompose the non-smooth differential pressure sequence into multiple smooth components, and screened the associated features by gray correlation analysis, and combined with a bidirectional long- and short-term memory network to achieve high-precision prediction. The experimental results show that the prediction error of this method for B5 batteries is controlled in the range of -1.87% to 1.43%, and the MAE, RMSE and MAPE indexes are reduced by 0.0081, 0.011, and 0.0122, respectively, compared with the LSTM method alone. This study provides a reliable health state monitoring technology for battery management systems, which is of great significance for extending the service life of the batteries and guaranteeing the safe operation of electric vehicles.
