Dynamic Bayesian network risk probability evolution for third-party damage of natural gas pipelines

Abstract

Failure and leakage of natural gas pipelines can lead to serious ecological losses and casualties. Third-party damage has become an important cause of pipeline failure and leakage, which urgently needs an accurate risk assessment method to assess the risk. Conventional qualitative risk analysis methods can only point out the critical events of failure accidents but fails to predict the failure probability. This paper proposes a dynamic risk probability analysis method based on Dynamic Bayesian network (DBN), which is validated by a third-party damage case under uncertainty. First, human factors are taken as the main analysis object in the risk analysis, by which two subcategories of intentional and unintentional factors are classified. A complete risk factor analysis is performed by combining expert recommendations with the fault tree analysis method and developing a coupled model with the event sequence diagram. Second, in order to deal with the uncertainty of risk factors, the coupled model is mapped to a DBN model. The prior probabilities of the input DBN model are obtained by database, fuzzy set theory, and Dempster-Shafer evidence theory. Weibull distribution is applied to construct the probability transfer process between time segments, which better fits the characteristics of third-party disruptive factors in onshore pipelines. Finally, the practicality and advantages of the proposed method are demonstrated by a real case study, which identifies 6 critical events and predicts the probabilistic information in different time slices. Furthermore, the method predicts the probability of failure events and potential consequences by processing the time series information, and it is found that the probability of structural damage and explosion is higher than other consequences. In this way, some risk management countermeasures are proposed in a targeted manner. The results show that compared with the conventional BN model which only performs probabilistic inference once, the DBN model can perform temporal dynamic inference to achieve the prediction of failure probability, and it can effectively achieve the numerical prediction of risk failure probability.