A Novel Training Set Optimization Method for Lithium-Ion Battery Capacity Degradation Trajectory Prediction

• Published in: 2024 3rd International Conference on Energy and Electrical Power Systems (ICEEPS) pp. 473 - 478
• Main Authors: Xie, Muhua, Chen, Liqun, Pei, Yuntian, Li, Chunbo, Zhang, Mingji, Shen, Wenjing
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.07.2024
• Subjects: Batteries; Battery lifespan prediction; Capacity aging trajectory prediction; Degradation; Lithium-ion battery; Long short term memory; LSTM; Neural networks; Optimization methods; Power system stability; Predictive models; Training; Training data; Training set optimization; Trajectory
• DOI: 10.1109/ICEEPS62542.2024.10693161

Capturing the lifespan trajectory of lithium-ion batteries (LIBs) in the early stage is critical for the operation and maintenance of electric vehicles (EVs). The battery's early cycling stage provides limited information in the training phase. This study introduces a long short-term memory (LSTM) neural network model embedded with an innovative training set optimization method to forecast the future degradation trajectory. Firstly, we propose a four-step training set optimization strategy (FS-TSOS) to screen out the training subset most similar to the target test set from the original training set, which can improve the quality of training data and eliminate redundant training data. Secondly, the LSTM neural network model is established based on the optimized training set to predict the capacity degradation trajectory. Thirdly, the comparison experiment between the proposed strategy and other novel methods is conducted to verify the superiority. The experimental results show that the proposed model can effectively predict the state of health (SOH) and remaining useful life (RUL) of the target battery even on the most basic neural network model because the overall degradation trend and internal degradation mechanism of the battery are fully considered.