分布选择对D-优化法估值精度的影响研究

Effect of Distribution Choice on Reliability Evaluation of D-Optimality-Based Sensitivity Test

采用Monte Carlo模拟仿真方法,研究了分布选择对D-优化法参数估值精度的影响,进而分析了估值结果对火工品99.9%的可靠性外推结果的影响。仿真结果表明:在D-优化法感度试验中,无论样本的真实分布如何,选择错误的其它分布对均值的影响不大,尤其在样本量较小的情况下;同时,无论样本的真实分布如何,选择错误的其它分布对标准差的影响都较大,这就导致了估值结果对99.9%的可靠性外推结果的影响较大。综合分析仿真结果,得出在不了解样品感度分布时,应尽量选择正态分布或对数正态分布进行感度试验,从而对99.9%可靠性外推结果的的影响最小。

Based on Monte Carlo method, the effect of distribution choice on parameter estimating precision of D- optimality-based sensitivity test was studied, and the effect of evaluating value on extrapolation 99.9%,6 reliability result was also analyzed. The results showed that whatever the sample＇s true distribution was, the effect of distribution choice on the mean was little, especially for a small sample, while the effect on the standard deviation was great when the error other distribution was chosen, which induced higher effect of evaluating value on extrapolation 99.9% reliability result. It can be educed from general analysis results that when the sample＇s sensitivity distributing is not well known, the normal distribution or lognormal distribution should be chosen, the effect of which on extrapolation 99.9% reliability result is the least in D-optimality-based sensitivity test.