Research on Thermal Conductivity Prediction of Natural Ester Insulating Oil Based on Transfer Learning

This paper proposes a thermal conductivity prediction model for natural ester insulating oil based on transfer learning, aimed at predicting the thermal conductivity of triglycerides. First, molecular dynamics simulations were used to calculate the thermal conductivity of FR3 vegetable oil and mineral oil at different temperatures, laying a foundation for expanding the data set of triglycerides. Next, molecular descriptors of 10 fatty acid molecules were calculated, and key features related to thermal conductivity were identified. Based on this, two regression models, extreme gradient boosting (XGBoost) and random forest, were used to construct a thermal conductivity prediction model for fatty acids. Knowledge from this model was then transferred to the triglyceride prediction task through transfer learning. The final model achieved an R2 score of 0.97 on the test set, significantly enhancing stability and generalization capability. This study applies transfer learning, enabling high-accuracy predictions even with small sample sizes. The developed prediction model can be integrated with molecular design, providing a scientific basis for designing insulating oils with high thermal conductivity. This not only promotes the development and application of new materials but also facilitates the optimization and sustainable development of energy systems.