Probabilistic Forecasting for Battery Health Estimation Using CNN-LSTM on Augmented Multi-Operation Data

Ensuring the safety and reliability of electric vehicles and energy storage systems requires an effective evaluation of lithium-ion battery State of Health (SOH). Most current methods are constrained to point-based estimations and are carried out under specific operating conditions, rendering them inadequate for accurately estimating the state of batteries with different electrochemical properties and working conditions. To solve these challenges, a CNN-LSTM-based interval prediction model is designed in this study. Data augmentation techniques, including jitter, scaling and amplitude distortion techniques, are used in the SNL dataset to expand NMC and NCA battery samples to 1 0 0 0 cycles while ensuring the same length as LFP battery samples. A total of 23,000 samples from 23 batteries are selected to train and test the performance of the proposed model. The CNN-LSTM model is trained to learn SOH representations, and interval estimation is achieved by computing the mean and variance at the output layer. Experimental results demonstrate that our approach offers high accuracy and improved interpret ability across diverse scenarios.