关键词: 数据污染, 全局灵敏度分析, 稳健统计量, Sobol指数

Abstract: The variance-based method of global sensitivity analysis has been widely used in many fields.To solve the problem that the sample mean and variance were susceptible to outliers in the sample under data contamination,an improved Sobol' Monte Carlo simulation method was proposed which integrated the robust statistic methodology with double loop reordering approach.The sample median and Hodges-Lehmann were introduced to substitute the sample mean to estimate the location parameters,and the median absolute deviation and Shamos were introduced to substitute the variance to estimate the scale parameters.Factor priorization and factor mapping were realized under the proposed robust double loop reordering approach in the situation of data contamination or skewed distribution.Simulation results showed that the proposed method was robust yet also outlier-resistant when the proportion of outliers in the data did not exceed 29.3%.It was not only efficient when the data were normal distribution but also had a good performance when the data were contaminated or skewed distribution.

Key words: data contamination, global sensitivity analysis, robust statistic, Sobol'indices