Remaining useful life prediction for lithium-ion batteries in highway electromechanical equipment based on feature-encoded LSTM-CNN network

Abstract

Lithium-ion batteries are the preferred choice for primary or emergency power supply in highway electromechanical equipment at present. The safety and longevity of these systems are significantly dependent on the health of the lithium-ion batteries. Accurate remaining useful life (RUL) prediction of lithium-ion batteries is essential for the reliable and continuous operation of highway electromechanical equipment. A RUL prediction method is proposed for lithium-ion batteries based on a feature-encoded LSTM-CNN network. Initially, statistical features are extracted by fitting a Gaussian mixture distribution to the probability density curve of the incremental capacity (IC) curve. Subsequently, physical features are derived from the battery discharge cycles. Statistical and physical features are then preprocessed, integrated and used in a feature-encoded LSTM-CNN network for RUL prediction. Testing on the NASA battery dataset demonstrated that this method reduces mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) by 31.0 %, 29.4 %, and 31.4 %, respectively, compared to traditional LSTM-based prediction models. The average error in RUL prediction is controlled within 1 cycle. Results validate that the proposed method has high precision and generalizability in predicting the RUL.