A novel method for variability analysis of model output

作者：Ren Bo^*; Cui Li jie; Li Ze

来源：IEEE Chinese Guidance, Navigation and Control Conference (CGNCC), 2016-08-12 to 2016-08-14.

Importance analysis is aimed at finding the contributions of the inputs to the output uncertainty. For this purpose, Sobol [Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates. 2001] puts forward a popular sensitivity measure based on the variance decomposition of the HMDR to evaluate the influence of the input uncertainty on the uncertainty of output. However, The main problem of decomposing the variance contribution based on HMDR[5] is not clear to some extent since the contribution of the individual input itself may include others' contribution. So, a novel variability measure and computational methodology of the variance contribution of the inputs are derived in this paper based on the RP-HMDR. Such is the main merit, is that the variance contribution of the individual input is estimated purely by other inputs being fixed to arbitrary values within their uncertainty ranges, which can reduce their effect in the output variance. Additionally, this paper explores the contribution from inputs in presence of the aleatory and epistemic uncertainties to output clearly by introducing a general auxiliary variable. Simultaneously, the contribution of the aleatory and epistemic uncertainties can be separated completely. Numerical and engineering tests show that the analysis of the variability contribution of the inputs can provide useful information for exploring the origins of the output variability, as well as identifying the structure of the model function for the complicated model with inputs in presence of the aleatory and epistemic uncertainties.

• 出版日期2016
• 单位中国人民解放军空军工程大学