一种贝叶斯诊断网络的拓扑结构

Topological Model for Bayesian Diagnostic Network

为了解决工程诊断中的不确定性,提出了一种基于工况-操作-故障-征兆的3层拓扑结构的贝叶斯诊断网络模型.该模型发挥了贝叶斯网络解决不确定性问题的优越性,融合了专家的经验知识,用图形化方式直观地表达了诊断中的复杂关系.应用实例表明,与传统质朴型贝叶斯诊断网络相比,该模型考虑了诊断对象的实际运行工况和操作情况,用于诊断的证据信息更充分,网络的诊断推理结果更符合诊断实际,已成功地应用于某石化公司炼油厂的烟机诊断中.

In order to solve the uncertainty problems in engineering diagnosis, a threelayered topological model using Bayesian diagnostic network was proposed based on condition, operation, fault, and symptom. Moreover, it used the superiority of Bayesian network in dealing with uncertainty and experts synthetically experience. The model represented the complicated relationships in engineering diagnosis intuitively. A practical example shows that its evidences for diagnosis are more sufficient and its result of network inference is more consistent with practice in comparison with naive Bayesian networks, because it takes account of operating actual conditions and operation records of equipment. This model has been used successfully for diagnosis of an industrial gas turbine in a refinery of a petrochemical complex.