10 - Safety and Security
A METHOD FOR ESTIMATING THE PROBABILITY OF RARE ACCIDENTS IN COMPLEX SYSTEMS
I. Romani de Oliveira¹, J.D. Musiak¹, G.C. Balvedi¹; ¹Boeing Research & Technology, Brazil
Rare event probability estimation of complex stochastic hybrid systems is a tough computational task. This type of analysis can be performed with sequential Monte Carlo approaches, but system modes with small probability of occurrence may be neglected if a proper importance sampling is not in place. This work presents an algorithm using a general-purpose optimization technique, denominated DIviding RECTangles, in order to find the combinations which mostly contribute to event occurrence. One of the by-products of this algorithms, a partition of the search space, is fundamental for the probability estimation. The theoretical framework used refers to Multi-Agent Dynamic Risk Models (MA-DRM) being mathematically expressed as General Stochastic Hybrid Systems, which allows modelling human behavior and interactions, besides technical elements, hence the denomination of “socio-technical”. On this basis, this paper systematizes methodological steps of analysis and applies them in a case study involving aircraft control. The approach proposed here is capable of obtaining statistically significant results for probabilities lower than any other method used for similar system models that could be found in literature, besides allowing the identification of intermediate events and making easier the evaluation of the sensitivity of the estimated probability to certain system parameters