Abstract: Addressing the complexities of power battery conditions, unpredictable surface temperature changes, and significant time lags encountered during the operation of heavy hydrogen fuel cards, this study focuses on the outside surface temperature of lithium-ion power batteries as the primary research target. To this end, an enhanced gate recurrent unit (GRU) neural network is proposed, optimized through the integration of a cross-entropy loss function and adaptive moment estimation (Adam). This approach establishes a surface temperature prediction model for lithium-ion power batteries, aiming to improve the accuracy and reliability of temperature predictions. The model uses the special gate mechanism and global processing ability of GRU neural network to obtain the nonlinear relationship between the surface temperature of lithium-ion battery and battery charging and discharging current, voltage, charging and discharging time, historical temperature, current temperature and ambient temperature. In this paper, four accuracy evaluation functions are used to evaluate the prediction model. The accuracy of the model is verified by simulation experiments under five ambient temperatures. The results show that the error of battery temperature prediction model based on GRU is relatively small compared with back propagation (BP) neural network model and recurrent neural network (RNN) neural network model, which indicates that the temperature prediction model of lithium-ion battery based on GRU has higher accuracy. This paper presents a new method for accurately predicting the surface temperature of lithium-ion phosphate battery.