Battery Core Temperature Sensing Using Electrochemical Impedance Spectroscopy and Data-Driven Regression: A Comparative Evaluation

Accurate monitoring of battery temperature is crucial for ensuring safety and optimal performance, particularly due to the limitations of traditional sensors, which measure only surface temperature. This study proposes a novel approach, utilizing Electrochemical Impedance Spectroscopy (EIS) as a sensing technique combined with Machine Learning (ML) to estimate core battery temperature. Following data acquisition and preprocessing, multiple ML models were trained and hyperparameters tuned via nested RandomizedSearchCV to ensure unbiased performance. A Random Forest achieved the best performance (RMSE $\approx 3.88^{\circ} \mathrm{C}$). To enhance interpretability, feature analysis revealed that the imaginary component of the impedance at specific frequencies contributes most to temperature estimation. The results highlight the potential of data-driven EIS analysis as a non-invasive alternative to surface sensors, paving the way for enhanced battery management systems (BMS).