Probability Analysis of Construction Risk based on Noisy-or Gate Bayesian Networks

Abstract

During construction risks’ probability assessment, it is challenging to obtain the joint probability distribution (JPD) of target risk systems, because every risk element's probability needs to be determined, known as the curse of dimensionality. This paper introduces a Noisy-or Gate Bayesian Network (NG-BN) model that integrates the Noisy-or Gate (NG) model and the Naive Bayesian Network (NBN) to address the problem. The NBN and the NG model's conditional independence assumptions’ gap is bridged by the Markov property. The proposed model requires only connection probabilities with high availability and reliability as the prior knowledge, thus substantially reduces the dimensionality of risk factors while retaining the ability of JPD reasoning. The model is illustrated and tested by a data analysis of the Zijingang Station construction project of Hangzhou Metro Line 5. The result demonstrates that the NG-BN can effectively accomplish the practical occurrence probability evaluation of construction risks. This study has a theoretical contribution as this model establishes a qualitative examination criterion of the Markov property. The proposed NG-BN performs better than the NBN on dimensionality reduction without diminishing the effectiveness of practical risk probability assessment. Its potential for reliability analysis in other engineering fields awaits further study.