A prediction model of insulation strength for gaseous medium considering the effect of external electric field

Context

To improve the prediction model of insulation strength for gaseous medium, it is needed to investigate the effect of external electric field on molecular microscopic descriptors. In this study, the global and local descriptors in the present of the external electric field are analyzed for non-polar gases and polar gases. According to the correlation analysis between molecular microscopic descriptors and insulation strength, both traditional regression and machine learning models can be used to predict the insulation strength of gaseous medium. The accuracy of insulation strength prediction models is effectively improved after considering the impact of external electric field on microscopic descriptors. The model based on the random forest achieves the highest accuracy. Furthermore after 1,000 rounds of training, the average R², MSE, MAE and NMBE of the test sets in the random forest model are 0.9239, 0.0346, 0.1581 and 0.1750, respectively. The average cross-validation score is 0.160, which is based on MSE as the evaluation criterion.

Methods

The Gaussian 16 software is utilized to optimize the 71 gas molecules using the M06-2X method and the 6–311 + + G(d, p) basis set. Molecular local descriptors are obtained using the wavefunction analysis software Multiwfn.