Lithium Battery Degradation Prediction Model Based on Capacity

Lithium battery degradation prediction has received wide interest in battery management systems. Battery degradation also indirectly helps improve the deployment of electric vehicles. This paper pioneers a data-driven battery degradation prediction model based on capacity in combination with modified ensemble empirical mode decomposition, mean impact value (MIV) and bidirectional long and short-term memory (Bi-LSTM) neural network. The proposed model exhibits superior degradation performance and improves prediction accuracy. First, the original capacity data are decomposed into intrinsic mode functions (IMFs) by MEEMD, and then MIV is added to reduce the IMF dimensionality. Second, the remaining IMFs are used as input to the Bi-LSTM neural network. Then, the degradation value is predicted in the next step. The proposed model is trained and tested on two large datasets. In the analysis of the autoregressive degradation model experiment, the proposed model achieves 0.0110 mean absolute percentage error and 0.0143 mean absolute percentage error on the Ames Research Center dataset. This work proves the feasibility and benefits of using the proposed model and also highlights how feature extraction can be used to improve predictions. According to the analysis of the influence of internal parameters on battery degradation, the role of measured voltage, measured current, and temperature in degradation prediction is about 1/3 on the Ames Research Center dataset. Internal parameters are proven to help predict lithium battery capacity in the case of limited access to battery capacity data hindering degradation prediction based on capacity, bringing a new perspective to degradation prediction.