管道腐蚀泄漏火灾的贝叶斯网络推理模型研究

Study on inference model of pipelines corrosion leak fire based on Bayesian networks

为了研究长输管道腐蚀泄漏及蒸气云爆炸事故的演化规律,通过对埋地管道内(外)壁腐蚀失效、燃气泄漏、气体云团扩散及蒸气云爆炸等4阶段事件进行分析,构建埋地管线腐蚀泄漏火灾的贝叶斯网络模型。研究网络结构中节点变量的取值范围及离散化方法,并基于对事故统计和专家分析判断,设定节点变量的先验概率,量化节点关联的条件概率分布。在对贝叶斯网络推理策略研究的基础上,考察节点变量对推理结果的敏感性,验证模型的合理性。结果表明,长输管道腐蚀泄漏及次生灾害事件过程具有较大的不确定性,主要体现在中间事件均具有多种状态,事故演化路径概率受模型输入条件影响较大。贝叶斯网络方法用于描述事故过程中间节点事件间的依赖关系有较大的优势,可以定量衡量事故风险的不确定性。

In order to research evolutionary laws of unconfined vapor cloud explosion（ UVCE） induced by combustible gas leak in long-distance oil and gas pipelines,Bayesian networks on buried pipelines corrosion leak fire were built by analyzing event nodes on inner and outer wall corrosion failure,combustible gas leak,the gas cloud diffusion and UVCE. The state ranges and discrete methods of node variables were studied. Priori probability and conditional probability distribution of the node variables were set by analyzing on accident statistics data and expert judgements. Bayesian network inference strategy was developed,the sensitivities of each network node variable on inference results were analyzed by researching on evolution mechanism of corrosion leak fire,and the rationality of the model was verified. The results show that there are greater uncertainty in the process of pipeline corrosion leaks and secondary disaster. The uncertainty presents in diverse intermediate event status value and probability of accident evolutionary path is influenced by the model input conditions. Bayesian network approach has a greater advantage to describe the dependency relations of accident intermediate nodes,and it can be used to measure uncertainties of accidents risk quantitatively.