Online multi-scenario impedance spectra generation for batteries based on small-sample learning

This is the authors' abstract. We don't add key points for this paper.

The onboard acquisition of data from electrochemical impedance spectroscopy (EIS) is critically important to the state assessment and fault diagnosis of mobile batteries, but it is technically challenging due to the stringent test requirements, limited modeling data, and varying mechanisms among batteries with different chemistries. This paper, without requiring any additional sensors, extends the traditional EIS measurement to online generation and covers most battery-using scenarios, including different battery chemistries, aging degrees, remaining capacities, and temperatures. Virtual simulation and transfer techniques are employed to train a deep neural network with a significantly reduced dataset. Specifically, we train the network with no more than 24 groups of data and achieve an average relative error lower than 5%, outperforming most "big data"-involved algorithms of its kind. Our method lowers the threshold of using EIS onboard and unlocks new opportunities to monitor the battery's performance in both time and frequency domain comprehensively in real time.