With the increasing voltage of ester oil-filled transformers and their application in offshore wind power

developing high-performance

fire-resistant esters is crucial for enhancing transformer safety. Flash point

a key indicator of fire-resistant performance for esters

is directly related to the molecular structures. This paper employs quantitative structure–properties relationship analysis to study the correlation between molecular structures and flash points of esters. Firstly

a database was constructed

comprising flash points and molecular descriptors of esters. Four molecular descriptors were selected using the correlation coefficients. Secondly

the predictive performance of models utilising linear regression

support vector machine regression and backpropagation neural network methods was compared. Finally

the model was used to assist in screening esters with high flash points and testing their performance as insulating esters. The results indicate that the neural network model with the tanh activation function provided the best prediction for the flash point of esters

achieving a coefficient of determination (R2) of 0.988 and a root mean squared error of 10.81°C for the test set. The predicted and measured flash points of the designed esters indicate that T2-C18:1 and T3-C18:1 achieved high flash points (> 250°C) and high breakdown voltages. This research study offers a novel approach to developing high-performance esters.