A case study on global sensitivity analysis with dependent inputs: The natural gas transmission model

This paper addresses the identification of the most important input parameters in a natural gas transmission model, in particular regarding their possible effects on pressure and temperature drops. This model has the peculiarity that a significant number of its uncertain input parameters are dependent on each other. Combinations of input parameters considered a priori as valid deliver impossible physical results (i.e.: negative pressures). This advises the application of a sampling method that rejects samples that lead to non-physical results. In a Bayesian framework, selective sample rejection modifies the a priori probability density function (pdf) of independent input parameters producing an a posteriori pdf with dependent inputs. Borgonovo's δ has been the Global Sensitivity Analysis measure selected for performing the sensitivity analysis. The results obtained are completely in line with what physical intuition indicates.