Symmetric approach to industrial safety risk assessment based on mutual probability correspondence

Abstract

The object of research is the process of assessing the level of safety of complex technical systems of critical infrastructure under conditions of uncertainty. The problem of the limitations and asymmetry of risk assessment methods was investigated. Risk assessment processes were studied based on IAEA data, using a combination of theoretical and computational modeling methods. The theoretical basis was based on factor risk analysis. Dynamic and temporal dependencies were taken into account using a synthesized modular scalable dynamic Bayesian network (MSDBN), which integrated local components and their interaction into hierarchical models. Probabilistic assessments were performed using Monte Carlo simulation, as well as structural and hybrid learning algorithms for Bayesian networks. The limitations, asymmetry, and dependence on expert opinion of traditional risk assessment methods were shown. It was shown that the synthesis of Bayesian networks and the Monte Carlo method as basic approaches meets the criteria for symmetry in risk event modeling. It was established that the maximum adequacy of risk event prediction is achieved when using a modular Bayesian architecture with a multi-criteria approach through assessing the compliance of production system elements with regulatory requirements, historical analogies and/or modeling results. MSDBN improves the quality and validity of management decisions, is integrated into automated control systems, serves as a tool for digital twins, can be used in the educational process, is symmetric and suitable for assessing the effectiveness of security measures. The proposed approach is useful for state, defense and industrial systems, including under conditions of uncertainty.