基于动态贝叶斯网络的邻近下穿隧道深基坑施工风险分析

为了减小邻近既有下穿隧道深基坑施工风险及灾害损失，科学预防施工安全事故，提出了一种基于动态贝叶斯网络（DBN）的深基坑施工风险分析模型。首先，运用BWM（best worst method）确定准则的权重;其次，基于关联规则挖掘风险因素间的相互关系，并以此构建DBN结构模型;最后，以新建厦门北站地下一层社会连廊深基坑工程为例，对提出的方法进行有效性和适用性检验。结果表明：基坑围护结构的安全度在静态被评为“较高”和“极高”的概率分别为34.6%和36.1%，且此结果随着输入风险证据发生动态变化，运用反向推理也能迅速找出围护桩渗水风险；提出的模型能明确邻近既有下穿隧道深基坑施工风险传递过程中的关键风险点，并能进行动态风险预测以及事故后致因诊断,从而实现邻近既有下穿隧道深基坑施工风险的动态管控。所提出的优化DBN模型对工前风险评估、先验分析和风险诊断有较好的适用性和较高的准确性，可为邻近既有隧道深基坑施工过程中的安全管控提供有效的决策支持，大幅提高风险控制效率。

Risk analysis of deep foundation pit construction in adjacent underpass tunnels based on dynamic Bayesian network

In order to reduce the construction risk and disaster loss of deep foundation pits in adjacent existing underpass tunnels and scientifically prevent construction safety accidents, a deep foundation pit construction risk analysis model based on dynamic Bayesian network (DBN) was proposed. Firstly, the BWM (best worst method) was applied to determine the weights of the criteria. Secondly, the interrelationships between risk factors were mined based on association rules, and the dynamic Bayesian network structure model was constructed in this way. Finally, the effectiveness and applicability of the proposed method were examined by taking the deep foundation pit project of the underground layer of the social corridor of the new Xiamen North Station as an example. The results show that the probability of the safety of the pit enclosure being rated as “high” and “very high” in the static state is 34.6% and 36.1%, respectively, and this result changes dynamically with the input risk evidence, and the risk of water seepage of the enclosure piles can be identified quickly by using reverse reasoning. The proposed model can clarify the key risk points in the risk transfer process of deep foundation pit construction in adjacent existing underpass tunnels, and can make dynamic risk prediction and post-accident causation diagnosis of the risk, so as to realize the dynamic control of the construction risk of foundation pit construction in adjacent existing underpass tunnels. The proposed optimized dynamic Bayesian network model has good applicability and high accuracy for pre-construction risk assessment, a priori analysis and risk diagnosis, which can provide effective decision support for safety control during the construction of deep foundation pits of adjacent existing tunnels, and greatly improve the efficiency of risk control.