An SOH Estimation Method Based on Real-World Vehicle Data from a Big Data Platform

Lithium-ion batteries equire accurate estimation of their state of health (SOH) to ensure safety and prolong their lifespan. We propose an innovative SOH estimation method based on real-world vehicle data. Specifically, battery capacity is calculated using high-current stable segments during the charging process, and the capacity degradation trend is qualitatively characterized using locally-weighted scatterplot smoothing, with health features extracted from historical operational data. Then spearman correlation analysis was used, correlation statistics of the features was calculated, and the dimensionality reduction of the health features was performed by principal component analysis, and a gated recurrent unit with attention mechanism model based on the SOH estimation of real vehicles was established. To evaluate the generalization capability of the proposed model, three real-world datasets from the same vehicle type but with different battery cells were selected.. The model was trained on one dataset and directly tested on the remaining two. Experimental results show that the maximum relative error was 4.7%, and, after curve fitting, the maximum relative error was reduced to 2.5%. Despite the differences in battery cell types, the model maintained high prediction accuracy, demonstrating strong adaptability and robustness to inter-cell performance heterogeneity. This method shows promise in providing effective support for optimizing battery management platforms