Bayesian Monte Carlo simulation-driven approach for construction schedule risk inference

As the construction of infrastructures becomes increasingly complex, it has often been challenged by construction delay with enormous losses. The delivery of complex infrastructures provides a rich source of data for new opportunities to understand and address schedule issues. Based on these data, many efforts have been made to identify key construction schedule risks and predict the probability of risk occurrence. Bayesian network is one of the most useful tools for risk inference. However, there are still two obstacles preventing the Bayesian network from being adopted popularly in construction schedule risk management: (1) the development of directed acyclic graph (DAG) and associated conditional probability tables (CPTs); and (2) the lack of observation data to trigger risk inference as evidence at the planning stage. This research aims to develop a novel Bayesian Monte Carlo simulation-driven approach for construction schedule risk inference of infrastructures, where the Bayesian network model can be developed in a more convenient way and be used without observation data required. It first constructs the key risk network with key risks and links through network theory-based analysis. Then the DAG structure of a Bayesian network is developed based on the topological structure of key risk network using deep-first search (DFS) and adapted maximum-weight spanning tree (A-MWST) algorithms. The CPTs are further developed using the leaky-MAX model. Finally, the Bayesian Monte Carlo simulation-driven risk inference method is developed for predicting and quantifying the probability of construction schedule risk occurrence. A real infrastructure project was selected as a case study to verify this developed approach. The results show that the developed approach is more appropriate to deal with risk inference of infrastructures considering its reliability, convenience, and flexibility. This research contributes a new way to construction schedule risk management and provides a novel approach for quantifying and predicting risk occurrence probability.