Developing an accurate equivalent electrochemical impedance-thermal coupling model is crucial for calculating impedance and estimating temperature during low-temperature heating of batteries using high-frequency alternating current (AC). The NSGA-II algorithm was used to identify parameters for seven equivalent impedance models across a frequency range of 10 Hz to 100 000 Hz. The study found that a model using a resistor and inductor parallel module better accounts for the skin effect at high frequencies without increasing computational complexity compared to a single inductor module. The proposed model, which incorporates time-varying heat transfer coefficients and high-frequency additional heat, was validated under both constant-frequency and variable-frequency conditions. It reduced the maximum temperature prediction error from 2.93 °C to 0.35 °C, with an RMSE of only 0.23 °C compared with existing models, proving its practicality and accuracy.