Robust SOH estimation for Li-ion battery packs of real-world electric buses with charging segments

Accurate estimation of battery packs state of health (SOH) is essential for the timely maintenance and efficient reuse of batteries in pure electric buses, which paly an import role in modern public transportation. However, real-world SOH estimation faces significant challenges due to inconsistencies among cells and variations in charge-discharge depths. Based on extensive operational data from electric buses, a novel SOH labeling calibration method is proposed, forming the foundation of a robust SOH estimation framework. First, the SOH of the battery pack is labeled using a variant of the ampere-hour integral formula applied to charging data, enhanced by mean filtering over 30 consecutive charge-discharge cycles to mitigate error influence. The charging data is then resampled to standardize segment lengths and fixed to 100 time-domain sampling points. Next, a convolutional neural network is employed to automatically extract features from the charging data, while a neural network maps these features to the battery pack SOH. Unlike existing methods, the proposed method accommodates charging sessions of varying durations and enables robust SOH estimation for electric buses battery packs without the need for manual health feature extraction. The proposed method is validated using approximately four years of operational data from three different types of electric buses. Through cross-validation, the method demonstrates high accuracy, achieving absolute errors below 3% in over 80% of cycle cases. The overall mean absolute error for SOH estimation across nine buses of three types is 1.585%, confirming its robustness and applicability.