Wang, F (Wang,Fan); Ding, LY (Ding, L. Y.); Luo, HB (Luo, H. B.); Love, PED (Love, Peter E.D.)

下载地址：

http://www.sciencedirect.com/science/article/pii/S0957417413004570

Abstract：There is anintrinsic risk associated with tunnel construction, particularly in urban areaswhere a number of third party persons and properties are involved. Due to thelimited availability of data for accidents and the complexity associated withtheir causation, it is therefore necessary to combine available historical dataand expert judgment to consider all relevant factors to undertake a realisticrisk analysis. Thus, this paper presents a hybrid approach that can be used toundertake a probabilistic risk assessment of the risks associated withtunneling and its likelihood to damage to existing properties using thetechniques of Bayesian Networks (BN) and a Relevance Vector Machine (RVM). Acausal framework that integrates the techniques is also proposed to facilitatethe development of the proposed model. The developed risk model is applied to areal tunnel construction project in Wuhan, China. The results derived from theproject demonstrated the model’s ability to accurately assess risks during tunneling,specifically the identification of accident scenarios and the quantification ofthe probability and severity of possible accidents. The potential of this riskmodel to be used as a decision-making support tool was also explored.

Keywords:Probabilisticrisk assessment; Bayesian networks; Relevance vector machine; Tunneling