A recovery model for battery data in the cloud and its application of state-of-charge estimation in electric vehicles

High-quality historical battery data is crucial for state estimation and management. However, due to limitations in network bandwidth and storage capacity, the cloud only receives low-frequency (LF) data, while high-frequency (HF) data is stored locally on the EV for a short period. This paper introduces a model-based framework for recovering low-frequency voltage signals. The training and test datasets are first constructed using real-world vehicle data. Subsequently, a multitask learning model within a semi-supervised learning framework is proposed to capture the HF voltage representation of each battery cell. The model successfully upsamples the 0.1 Hz sampling rate data to 1 Hz with a root mean square error (RMSE) of 16.71 mV on the test dataset post-training. A SOC estimation framework based on the unscented Kalman filter and electrochemical model is introduced to capitalize on the high-quality data generated by the voltage recovery framework. This framework estimates SOCs for each individual cell in a battery pack by identifying unique electrochemical parameter sets for each cell. The results demonstrate that the framework can identify the cell with the lowest real SOC and estimate SOC within a 1.8% RMSE margin, even when the cell with the lowest SOC does not exhibit the lowest voltage. Finally, cost-effective enhancements for both the voltage recovery model and the SOC estimation framework are recommended to balance performance with the computational power requirements.