Showing papers presented at "Conference on Scientific Computing in 2014"

Comparison of interval and Monte Carlo simulation for uncertainty propagation in atmospheric dispersion model

Abstract: In this paper, the problem of tackling uncertainty propagation in the estimation of the atmospheric dispersion of a toxic gas release is analyzed in order to assess the risk at the event of an accident. This estimation is based on an effect model associated with the studied dangerous phenomenon where some input variables and model parameters are known with imprecision. Two simulation approaches, Monte Carlo and interval analysis method, are applied and compared for estimating the confidence interval of risk intensity. Interval analysis method is superior in estimating all the possible values of intensity relative to the Monte Carlo simulation. A sensitivity analysis based on Sobol indices is applied in order to reduce the number of uncertain variables while conserving an acceptable precision of effect model. Furthermore, much less computational time is required for interval analysis method than for Monte Carlo simulation.

Tuning the accuracy of rational macromodels to nominal load conditions

Abstract: We address the generation of broadband macromodels of complex linear systems via rational curve fitting. We show that standard approaches may not ensure that the macromodel accuracy is preserved in system-level simulations, under loading conditions that are different from the adopted identification settings. Our main contribution is an automated procedure for the definition of a frequency-dependent norm weighting strategy that tunes the macromodel accuracy for a specific nominal termination network, thus improving model robustness under realistic operation