| 摘要: |
| 跨水桥梁在水文环境和多重荷载耦合作用下，其运行期风险因素间存在复杂的关联关系。传统风险评估方法难以有效挖掘风险因素间的深层因果逻辑与传递路径，为此，本文提出一种融合Apriori关联规则、ISM解释结构和贝叶斯网络的协同分析框架，系统解析跨水桥梁运行期风险致因链并预测风险率。首先，基于跨水桥梁事故案例数据构建风险因素、失效模式与事故状态的二值化矩阵，运用Apriori算法挖掘风险因素与失效模式间的强关联规则。其次，结合ISM模型对关联规则进行层级化分解，将风险因素划分为根源层、中间致因层、失效模式层和结果层，揭示风险从源头到结果的层级传递逻辑，进而构建跨水桥梁的风险贝叶斯网络。结果表明，该模型具有良好的预测精度和稳定性，预测准确率达到88.1%，AUC值为0.852；暴雨洪水与养护缺失是对桥梁倒塌影响最显著的根源因素，河床冲刷下切和材料腐蚀劣化是关键的中间致因因素；暴雨/洪水→河床冲刷下切→基础淘空/墩台倾斜→倒塌是跨水桥梁倒塌风险最为突出的传递路径；研究结果可为跨水桥梁的安全管理提供决策支持，提升风险防控能力。 |
| 关键词: 跨水桥梁 关联规则 解释结构模型 贝叶斯网络 风险致因链 风险率 |
| DOI： |
| 分类号: |
| 基金项目:甘肃省交通运输厅科技项目(2025-35) |
|
| Analysis of Risk Causation Chain for Water-Crossing Bridges During Operation Based on Apriori-ISM-BN Framework |
|
ZHU Haofeng1, ZHOU Zheng1, LOU Benxing2, HU Wenzhe2
|
|
1.Gansu Luqiao Highway Investment Co,Ltd;2.Yellow River Engineering Consulting Co,Ltd
|
| Abstract: |
| Under the coupling effects of hydrological environment and multiple loads, the risk factors of water-crossing bridges exhibit complex interrelationships during operation. Traditional risk assessment methods are difficult to excavate the deep causal logic and transmission pathways among risk factors. To address this limitation, this paper proposes a collaborative analysis framework integrating Apriori association rule mining, Interpretive Structural Modeling, and Bayesian Network to systematically analyze the risk causation chain of water-crossing bridges during operation and predict the failure probability. The proposed framework leverages the respective strengths of the three methods. The Apriori algorithm is employed to objectively mine strong association rules from a large volume of accident case data, thereby overcoming the subjective bias inherent in purely expert-dependent approaches. ISM is then applied to decompose the extracted association rules into a hierarchical structure, explicitly revealing the direction and logical sequence of risk transmission from root causes to surface outcomes. Finally, a Bayesian network is constructed based on the ISM-derived structure to enable quantitative prediction of risk probabilities and scenario-based inference under various evidence conditions.
A total of 718 historical accident cases of water-crossing bridges worldwide, spanning from 1950 to 2024, are collected and analyzed, encompassing multiple bridge types including girder bridges, arch bridges, cable-stayed bridges, and suspension bridges. The risk factors are categorized into 15 types, failure modes into 6 types, and the bridge state is defined as the final response variable. To evaluate the predictive performance and generalization capability of the constructed Bayesian network, a dual validation strategy is adopted. A set of 68 independent cases is randomly held out as a validation set, which does not participate in any modeling process and is used solely for final generalization assessment. The remaining 650 cases are subjected to ten-fold cross-validation, and the average performance is reported. Sensitivity analysis is further conducted using mutual information to identify the most influential risk factors for each failure mode, providing prioritized guidance for risk prevention and control. The most critical risk causation chains are extracted by calculating the joint contribution and conditional probability of each transmission path. The results demonstrate that the proposed model achieves good predictive accuracy and stability. Under the optimal threshold combination of 1% minimum support and 10% minimum confidence, the model attains an accuracy of 88.1% and an AUC value of 0.852 in ten-fold cross-validation. On the independent validation set, the model achieves an accuracy of 86.8%, a recall of 82.4%, and an AUC of 0.841, indicating satisfactory generalization capability without significant overfitting. The results further reveal that rainstorm and flood, as well as maintenance deficiency, are the most significant root factors affecting bridge collapse, while riverbed scour and material corrosion are identified as the key intermediate causal factors. The most prominent risk transmission path for bridge collapse is found to be rainstorm and flood → riverbed scour → foundation scour or lateral displacement → collapse, with a joint contribution of 0.128 and a conditional probability of 0.290. Another critical pathway is maintenance deficiency → material corrosion → girder displacement or damage → collapse, with a joint contribution of 0.077 and a conditional probability of 0.174. Ship and vehicle impact is also identified as an important external disturbance factor that can directly cause girder displacement or damage and pier tilting. The findings provide valuable decision support for the safety management of water-crossing bridges, enabling authorities to prioritize risk mitigation measures during flood seasons, strengthen scour monitoring and protection, enhance routine maintenance to slow material deterioration, and improve vessel navigation management and anti-collision facility construction. |
| Key words: Water-crossing bridges Apriori algorithm Interpretive structural modeling Bayesian network Risk causation chain Risk rate |
