Improved SOH Prediction of Lithium-Ion Batteries Based on Multi-Dimensional Feature Analysis and Transformer Framework

Data-driven state-of-health (SOH) prediction is increasingly critical for the effective management of lithium-ion batteries; however, challenges remain in practical applications. Traditional methods that rely on a single health indicator often fail to capture the complexity and multi-dimensional nature of battery performance changes. To address these limitations, this paper presents a novel Transformer-based approach for accurate SOH prediction. The correlation between various measured and computed features extracted from battery charge/discharge curves and their impact on battery performance degradation are investigated using Pearson correlation coefficients. Three strongly correlated features are identified as multiple input variables for the Transformer framework. The effectiveness of this Transformer-based SOH prediction method is demonstrated using public datasets, revealing that predictions for internal resistance and capacity closely align with actual values, with most RMSE values falling below 0.01. Furthermore, validation with an additional laboratory dataset confirms the accuracy and adaptability of our proposed approach, highlighting its potential to enhance SOH prediction in real-world applications.