A temperature prediction method for electric vehicle charging connectors using a CFD-based deep learning model

Abstract

Aiming to rapidly predict the temperature of electric vehicle (EV) charging connectors in dynamic environments and clarify the mechanisms by which various parameters affect temperature characteristics during the charging process. The reliability of the computational fluid dynamics (CFD) simulation was first validated through experiments. Subsequently, a systematic CFD analysis was conducted to investigate the effects of contact resistance, ambient temperature, and charging current on the internal temperature distribution of the charger. Additionally, a novel data-driven approach was proposed, utilizing an improved sparrow search algorithm (ISSA) to optimize a long short-term memory (LSTM) neural network for real-time temperature prediction. This approach enhances predictive performance through hyperparameter optimization. The results show that the CFD model has high accuracy, with a maximum error of 7.53 %. CFD analysis clarified the effects of various parameters on the charger's temperature rise. Furthermore, the proposed optimization strategy significantly improved model performance, with the ISSA-LSTM model achieving a 50 % reduction in mean absolute error compared to the conventional LSTM model. The model also demonstrated strong generalization capability, with the maximum absolute error remaining within 5 °C in test cases beyond the training data range. This method provides an effective and reliable tool for temperature prediction in EV charging systems, with considerable industrial application potential.