Uso de técnicas de metamodelagem para análise de sensibilidade e quantificação de incertezas na modelagem hidrodinâmica de perdas de vidas associadas a inundações tecnológicas

Abstract

Metamodeling is frequently used to propagate uncertainties, i.e., to determine output statistics of a mathematical or computational model considering the probability distributions of the input variables, and to perform sensitivity analysis of these variables. In this context, the objective of this study was to conduct a sensitivity and uncertainty analysis of the input variables in the computational model Lifesim, used for modeling loss of life stemming from natural and technological flood events. To achieve this, the hydrodynamic and loss of life modeling for the hypothetical rupture of the Pampulha Dam in Belo Horizonte - MG was performed. The methodology of the study is divided into three stages. In the first stage, a two-dimensional hydrodynamic model was updated using HEC-RAS v. 6.3 software. The second stage involved deterministic scenarios considering efficient and inefficient alert and evacuation systems for both daytime and nighttime. In the third stage, a substitute model (metamodel), based on Polynomial Chaos Expansion, was constructed using MatLab software. Overall, the results of the substitute model were appropriate. The error of the Polynomial Chaos Expansion ranged from 10^(-3) to 10^(-2) for the simulated scenarios, and the mean squared error between the substitute model output and the computational model (LifeSim) resulted in differences of 1 to 2 fatalities. The global sensitivity analysis, calculated using the Sobol Index up to the third order, was conducted for the parameters "Alerted Population," "Mobilized Population," "Alert Emission," and "Hazard Identification." For all scenarios, the variables "Alert Emission" and "Hazard Identification" had the greatest contribution to the model output, i.e., to the number of fatalities. However, in the optimistic scenarios, which assume that the population is prepared for mobilization and evacuation and that the alert and evacuation system are efficient, the variable "Mobilized Population" presented more significant indices. In conclusion, the use of the substitute model based on Polynomial Chaos Expansion for uncertainty and sensitivity analysis of the input variables proved to be useful.