State of Health Estimation of Lithium-Ion Batteries Based on KAN Architecture with BiLSTM

With the rapid development of electric vehicles and renewable energy storage systems, the estimation of lithium-ion battery State of Health (SOH) has become increasingly important. Effective SOH monitoring not only extends battery lifespan but also enhances the overall efficiency and reliability of the system. This paper presents SOH method that combines the Kolmogorov-Arnold Network (KAN) with Bidirectional Long Short-Term Memory (BiLSTM) networks. First, features are extracted from historical data, resulting in 14 features. The features are then categorized into two types: one type consists of noisy features, and the other includes features with anomalous measurement data. The Variational Mode Decomposition (VMD) parameters are adjusted for each category, and the processed features are used as indirect features. Three different methods (LSTM BiLSTM and GRU) and three hybrid models (LSTM-KAN BiLSTM-KAN and GRU-KAN) are employed to establish SOH estimation models. The performance of the six models is compared and analyzed. Results indicate that models combined with the KAN network outperform their baseline counterparts, with the BiLSTMKAN model exhibiting the highest estimation accuracy compared to the other models. It achieves lower Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE), significantly enhancing prediction accuracy and efficiency.