Data-driven internal temperature estimation methods for sodium-ion battery using electrochemical impedance spectroscopy

As sodium-ion batteries (SIBs) move towards commercialization, safety monitoring of SIBs has become the next key issue, and how to avoid thermal runaway is one of the toughest challenges. The electrochemical impedance spectroscopy (EIS) method for the internal temperature estimation of lithium-ion batteries has received considerable attention due to its non-invasive detection and high accuracy. However, research on the impedance-temperature characteristics of commercial SIBs remains limited, and EIS methods suitable for estimating the internal temperature of SIBs require further investigation. In this study, four commercial 26,700 SIBs were tested, the EIS results of the batteries in various state-of-charge (SoC) states and at different temperatures were systematically investigated, and a method for estimating the internal temperature of SIBs on the basis of the combination of EIS and machine learning (ML) is proposed. Seven component parameters were extracted as features from the raw EIS data using equivalent circuit model fitting, and four features, which were found to be highly correlated with temperature, were further selected by correlation analysis. The mapping relationship between the extracted features and the internal temperature of the battery was established based on three ML regression models. Results demonstrate that the average estimation error of the multi-layer perceptron models for the internal temperature of the battery with unknown SoC states is only 1.086 °C. This paper fills the gap in the temperature characterization of EIS for SIBs and proposes an effective method for overcoming the cross-coupling of the battery EIS with the SoC and temperature within the framework of mechanical learning to estimate the internal temperature of SIBs.