模糊动态贝叶斯网络在深基坑施工风险演化分析中的应用

Application of fuzzy dynamic Bayesian network in risk evolution analysis of deep foundation pit construction

为解决深基坑施工过程中风险因素随时间变化的动态特征带来的施工风险演化问题,提出了一种基于模糊动态贝叶斯网络(Fuzzy Dynamic Bayesian Networks,FDBN)的深基坑施工风险分析方法。该方法从人、机、环、管、材料、技术6个方面提取及合并风险因素,并采用故障树对风险因素的因果关系加以表示,再将故障树映射为贝叶斯网络图形结构;采用模糊集理论和改进的相似聚合法(Similarity Aggregation Method,SAM)确定网络节点的先验概率和条件概率,基于马尔科夫链确定时间片间的转移概率,从而构建完整的FDBN模型预测施工风险概率的动态演变。将案例工程实时信息作为模型输入,结果表明:所建模型根据输入证据准确预测了工程风险概率变化;通过敏感性分析识别了关键风险因素并对其进行了排序。最后,结合理论验证进一步证实了该方法的有效性。

To solve the problem of risk evolution caused by the complex risk factors in deep foundation pit construction and the change of risk factor state with time,a deep foundation pit construction risk evolution analysis method based on fuzzy set theory and Bayesian network is proposed.Firstly,based on the statistical analysis of deep foundation pit engineering accidents in past years and literature research,the main factors leading to deep foundation pit construction risks were extracted and screened from six aspects:human,machine,environment,management,material,and technology.A fault tree model was used to represent the risk identification results.Secondly,according to the mapping rules of the fault tree model,the fault tree was first mapped into a static Bayesian network and the node states were determined,and then dynamic nodes were divided to establish a dynamic Bayesian network for risk evolution analysis.Finally,fuzzy set theory was used to convert expert evaluation opinions into exact values,and an improved similarity aggregation method was introduced to reduce the subjective influence of expert evaluation to determine the prior probability and conditional probability of nodes.The transfer probability between time slices was determined based on Markov chains.All probability parameters were input into the constructed graphical structure,and a complete Fuzzy Dynamic Bayesian Networks(FDBN) analysis model was constructed with the help of GENIE software.The model was used to analyze the case project,and the results show that the model accurately predicts the evolution trend of risk probability by taking the change of risk factor state as the evidence input model.With the help of the model's backward diagnostic capability and sensitivity analysis method,the top three key risk factors affecting the construction safety of deep foundation pit were identified as follows:instability of support structure,over-excavation of earthwork,and substandard strength of poured concrete.Finally,the validity of the method was further confirmed by combining it with theoretical verification.