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

Key words:  long-distance oil and gas pipeline  corrosion  leak  vapor cloud explosion  Bayesian network

Abstract:

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.