Battery SOH assessment for real-world EVs based on discharging process characteristic and ensemble learning approach

Most existing research estimates the battery state of health (SOH) of real-world electric vehicles (EVs) using charging data, while ignoring the information embedded in the discharging data from daily operations. To address this gap, this paper proposes a novel SOH estimation approach based on real-world driving data, augmented by an ensemble learning (EL) strategy. Specifically, the empirical mode decomposition method is employed to generate health indicators (HIs) from segments of interest, which are extracted from real-world discharging signals using a novel fragment separation technique. Then, considering the comprehensive and performance limitations of single models in complex real-world applications, an EL-based approach integrating two carefully discussed and selected base models is developed. Moreover, a mileage-based weight adjustment strategy is also proposed to compensate for divergent degradation trends caused by battery inconsistencies. Numerically, the proposed EL-based SOH estimator trained by HIs extracted from discharging datasets can reduce the average mean absolute error (MAE) by about 14% compared to two base models working alone, verifying the effectiveness of real-world driving signals and the superiority of the proposed EL strategy. The proposed weight adjustment method can also decrease the average MAE by about 19% compared to traditional weight updating methods.