부산대학교 도서관

Obtaining high-resolution battery historical data in the cloud is crucial for lithium-ion battery state estimation and thermal runaway prediction. Due to limited signal transmission bandwidth, the cloud primarily stores low-frequency (LF) and less high-frequency (HF) data. This article proposes a model-based data compression and recovery method to efficiently transfer battery signals between electric vehicles and the cloud. First, training datasets are generated from real vehicle data. Then, a multitask learning model, within a semisupervised framework, is presented to learn the HF voltage representation of each cell. The semisupervised learning task utilizes unlabeled LF voltages to enhance the voltage recovery effects. Multitask learning is employed to address the issue of target domain drifting caused by cell aging. Finally, battery data from vehicle and laboratory tests are utilized to compare the results of different methods on voltage recovery tasks. The results demonstrate that the proposed method can reduce the average voltage recovery error to less than 8 mV with a compression ratio of 10.