A Data-Driven DAE-CNN-BiLSTM-Attention Prediction Model for the State of Health of Lithium-Ion Batteries

Accurately predicting the health state of lithium-ion batteries is essential for their safety, reliability, and longevity. Predicting State of Health (SOH) using health indicators is a proven and effective method. However, real-world battery charge-discharge data is often noisy, particularly during capacity regeneration. To achieve accurate health state predictions, we extracted over ten health indicators and designed a hybrid model: DAE-CNN-BiLSTM-Attention. This model integrates the strengths of Convolutional Neural Networks (CNN) for local feature extraction, Bidirectional Long Short-Term Memory networks (BiLSTM) for temporal dependency learning, the Attention mechanism for effective weight assignment, and Denoising Autoencoders (DAE) for restoring original data, enabling the network to better adapt to complex real-world environments. The adaptability and stability of the proposed model were validated using two public datasets: NASA and CALCE. Compared to other existing methods, the proposed model demonstrated superior performance, achieving mean absolute error (MAE) and root mean square error (RMSE) of 0.0154 and 0.0191, respectively.