Compression and inference algorithms for Bayesian network modeling of infrastructure systems

Abstract

The Bayesian network (BN) is an ideal tool for modeling and assessing the reliability of civil infrastructure, particularly when the information about the system and its components is uncertain and evolves in time. One of the major limitations of the BN framework, however, is the size and complexity of the system that can be tractably modeled as a BN. This is due to the size of the conditional probability table (CPT) associated with the system node in the BN model, which grows exponentially with the number of components in the system. In this paper, we present novel compression and inference algorithms that utilize compression techniques to achieve significant savings in memory storage of the system CPT. In addition, heuristics developed to improve the computational efficiency of the algorithms are presented. An application to an example system demonstrates the gains in both memory and computation time requirements achieved by the proposed algorithms.