Vehicle-cloud collaboration method enables accurate battery state of health estimation under real-world driving conditions

Accurate and reliable State-of-Health (SOH) estimation is critical for battery electric vehicles (BEVs) battery systems to ensure safe and stable operation. Facing real-world driving conditions of BEVs, this work introduces a vehicle-cloud collaborative SOH estimation method with hybrid data-driven modeling. The proposed method addresses challenges like the poor modeling effect of conventional SOH data-driven estimation methods under the modeling data with probability distribution differences, as well as the limited learning capabilities of conventional data-driven algorithms, achieving both high-accuracy and real-time SOH estimation. Specifically, a statistical indicators-based online incremental aging features construction method is proposed to ensure the applicability of the constructed features. By introducing ΔSOH data-driven modeling, the SOH and ΔSOH collaborative hybrid data-driven modeling strategy is designed, so that the model based on different vehicle driving condition data with probability distribution differences can still obtain reliable SOH estimation results. A recursive gradient boosting modeling framework is designed to enhance learning ability of cloud-based data-driven algorithms, and a feature self-scaling and forgetting factor is added for real-time online learning stability of vehicle end's model. Finally, the designed vehicle-cloud collaborative SOH estimation strategy involves self-monitoring and model updating on each end as well as mutual monitoring between different ends to obtain stable and effective SOH estimation results. Experimental results from two real-world BEVs indicate that the constructed features are effective, the designed vehicle-cloud collaborative SOH estimation method can still obtain results with an SOH estimation error of less than 2.5 % across different vehicle data domains.