Reliability Assessment of Critical Infrastructure Using Bayesian Networks

Abstract

The authors present a Bayesian network (BN)-based approach for modeling and reliability assessment of infrastructure systems. The BN is a powerful framework that is able to account for uncertainties in component and system parameters, and perform updating of system assessments with new information. The exponential increase in memory storage required for the BN model as the size of the system increases has limited the applicability of BNs for reliability assessment of large infrastructure systems. Recently, a data-compression method was proposed to address this limitation. While significantly reducing the memory storage, computational time for constructing the BN and performing inference increased. In this paper, new methodologies are developed to increase the computational efficiency of a compression-based approach for BN modeling and reliability assessment of infrastructure systems. These include algorithms to improve the computational efficiency of the initial compression for constructing the BN, subsequent inference over the network, and overall system formulation. The algorithms are applied to a test example system to examine their performance for systems of increasing size, as well as to a 59-component power distribution network to demonstrate application to real systems. Performance of the proposed methodologies is compared to that of an existing, widely used BN algorithm. With the heuristics employed, the new algorithms are shown to achieve significant gains in both memory storage and computation time, enabling the modeling of large infrastructure systems as BNs for system reliability analysis.