环保绝缘气体介电强度预测方法评估

The Evaluation of Dielectric Strength Prediction Methods for Eco-Friendly Insulation Gases

为明确各类气体介电强度预测方法的准确性,基于9种气体(N_(2)、CO_(2)、N_(2)O、CF_(4)、SO_(2)F_(2)、c-C_(4)F_(8)、C_(3)F_(8)、C_5F_(10O)和C_(4)F_(7)N)在50 Hz工频交流电场下测得的相对介电强度对5种预测方法进行了评估。结果表明,各方法预测准确性与其给出的可决系数R~2不一致。这主要与其使用的样本数据测试条件不统一、同一气体的相对介电强度不一致有关。当修正上述问题后,各方法准确性得到提升,并与其R~2相符。分子参数的丰富和放电理论的介入可使预测方法准确性提升。使用样本外气体进行验证的方法,其预测结果在修正前后都具有较高的准确性;使用基团贡献法的预测方法也具有上述特性。通过上述评估结果可知,样本中的气体数据需在统一条件下获得,并根据试验条件和样本分子特征确定预测方法的适用范围。放电理论、分子几何特征和新分子参数的引入有助于提高预测模型的准确性。模型验证和基团贡献法可提高预测方法对样本数据的包容性。

SF_(6)has a strong greenhouse effect,with a global warming potential(GWP)of 23500.It needed to be alternated urgently.Gas dielectric strength prediction methods could quickly give the dielectric strength based on molecular parameters.The existing prediction methods are various.They use different parameters and discharge theories.The fitting samples they used were from different test conditions.That would lead to uniform dielectric strengths for the same gas in each perdition method.This paper evaluated the existing prediction methods to clarify the impact of the abovementioned factors on accuracy and reliability.According to the parameters and model forms,five prediction methods with typical characteristics were selected.Method 1 used linear models of molecular polarization characteristics and ionization,adsorption energy,and other parameters.Method 2 added molecular geometric characteristics.Method 3 was based on the method 2 adding molecular symmetry factors,and the model is no linearized considering the discharge theory.Method 4 was a linear model using general interaction properties function(GIPF)parameters.Method 5 was the group contribution method.Nine frequently occurring gases(N_(2),CO_(2),N_(2)O,CF_(4),SO_(2)F_(2),c-C_(4)F_(8),C_(3)F_(8),C_(5)F_(10O),and C_(4)F_(7)N)were selected to evaluate the prediction method.The relative dielectric strengths of nine gases relative to SF_(6)were tested under the same conditions(50 Hz alternative current uniform electric field).The results show that the prediction accuracy of each method was lower than the fitting results given in their papers.That was because the test conditions of the gases used in each prediction method were not uniform,and the relative dielectric strength of the gases used was inconsistent.Some gases’dielectric strength had significant difference with the test data in this paper.The unification of gas dielectric strength based the test results in this paper could improve the accuracy of methods and the predication results of 5 methods could be consistent with their fitting results.In addition,with the enrichment of molecular parameters and the intervention of discharge theory,the prediction accuracy of methods 1 to 4 gradually increased.In addition,method 3,validated by the out-of-sample data,had high accuracy before and after the unification.Method 5,using the group contribution,also had the above properties.However,the groups method 5 could predict was limited.Hence,the method 5 could not predict all of 9 gases’dielectric strength.From the above evaluation results,it could be found that the gas data in the samples for fitting prediction models should be obtained under the same test conditions(electrode,electric field,voltage form,pressure,temperature,gap,et al).In addition,it is necessary to determine the applicable conditions and objects of the prediction method according to the sample data's test conditions and molecular characteristics.Towards the same sample,introducing the discharge theory,molecular geometric features,and new molecular parameters could help improve the accuracy of the prediction model.In addition,using out-of-sample data to validate the model and introducing the group contribution method could improve the inclusiveness of the prediction method for the sample data.