Novel differential voltage features based machine learning approach to lithium-ion batteries SOH prediction at various C-rates

Lithium-ion batteries are widely employed in consumer electronics and electric vehicles. Accurate estimation of their state-of-health (SOH) is critical for monitoring performance and understanding degradation mechanisms, and various approaches to extracting critical features from real-time series data have been considered in existing studies. This study proposes a novel SOH prediction approach that integrates differential voltage (DV) features, extracted from DV curves, with machine learning models in addition to conventional features extracted from current-voltage and incremental capacity (IC) curves. Experimental tests were conducted in a controlled laboratory environment to collect degradation data from lithium-ion batteries under varying C-rates. Several machine learning models, including convolutional neural networks with long short-term memory (CNN-LSTM), LSTM, recurrent neural networks (RNN), and multilayer perceptron (MLP), were trained and evaluated using the extracted features. Comparative analyses revealed that incorporating DV features significantly improved SOH prediction accuracy across diverse C-rates. Among all models, the CNN-LSTM with DV features achieved the best performance. As an example, for Cell #2, it yielded a mean absolute percentage error of 0.56 %. This approach provides a new way for advancing battery health prognostics and developing more accurate battery management systems.