Revisiting driving factor influences on uncertain cascading disaster evolutions: From perspective of global sensitivity

Abstract

For the simulation of a cascading disaster (or failure), the quantitative relationship between the disaster level and the parameters that define the driving factors of the cascade dynamics assist in identifying the critical factors and their thresholds to suppress the disaster outbreak. The variation of the disaster level, as well as the synchronous influences of all parameters and their interactions on this variation, conditional on global uncertain parameters, have long been neglected in previous analysis paradigm. This study revisited these issues using a global sensitivity analysis framework based on the extended Fourier amplitude sensitivity test (EFAST). The demonstrative experiment on a disaster causality model demonstrated that the uniform distribution of each parameter with bounds at only ± 20% of their default could increase the disaster level by thousands times. However, the individual or interactive influence of a small fraction of the parameters dominated this variation, and the influence of each parameter was significantly time-varying and different from each other. The disaster evolution and parameter influence on each network component were significantly affected by the propagation structure and distance from the initial disturbance. A comparison with the past research paradigm verified the merit of the EFAST, which acquires the time-dependent influences of all parameters and their interactions in a synchronous manner. This capability makes this method superior to instruct interventions on disaster cascade by identifying the influential driving factors.