Establishment of Reduced-Order Models for Temperature Prediction in Electric Vehicle Charging Gun

Abstract With the rapid development of liquid-cooled charging technology, the thermal management of the charging gun of electric vehicles has also become an urgent challenge to be solved. In this article, an accurate simulation model is developed by correcting the thermal conductivity and heat generation of the model through the integration of experimental results. Subsequently, an experiment is designed with current, coolant inlet flow, coolant temperature, and ambient temperature as variables, from which a series of charging gun temperature simulation results are obtained. Utilizing these results as a training set, a reduced-order model for predicting charging gun temperature is established. The influence of the multilayer perceptron model, response surface model, and Gaussian process model on prediction accuracy is then compared. The results indicate that compared to the other two models, the multilayer perceptron model has a more significant advantage in fitting accuracy for nonlinear variables, with an average temperature prediction error of 1.61 °C. This study holds significant importance for the intelligent thermal management of charging devices.