安全与环境学报

高含硫集气站中,井口分离器风险分析是集气站安全运行管理的重要环节。传统的因果图风险评价分析方法具有简洁直观,逻辑性强的优点,但具有一定的局限性。贝叶斯网络是一种较新的系统风险分析方法,能较好地表达变量之间的不确性关系且具有双向不确定性推理能力,但不如前者形象直观。采用因果图和贝叶斯网络对分离器液位过低事故进行分析并对比,充分利用两者的优点。结果表明,利用两种方法可以更好地对分离器进行风险分析。

The present paper is aimed at introducing the combination of cause-effect diagram and Bayesian network in hoping to improve the risk assessment of the wellhead separator in high-sulfur natural gas gathering station.As is known,conventional evaluation methods,such as the cause-effect diagram,are likely to produce ineffective and inaccurate assessment,due to their limitations in assessing events binary states and necessary logical relationship of the problems involved,due to their being too simple in structure and intuitive in slow reaction.Since Bayesian network is a kind of rather new risk analysis method,it has made it possible to better express the uncertainty among the variables,reason out the two-way uncertainty,and describe multi-states of the events more clearly.Moreover,it can help to produce an image more accurately than before.This paper would like to analyze low liquid-level separator by cause-effect diagram via Bayesian network and then make a comparison between the results.What is more,we have carefully studied the advantages of the above-said methods.To be sure,first of all,we have simulated the accidents caused by the low liquid level of the separator via the cause-effect diagram.Next,we have tested and analyzed the wellhead separator security barrier failure and the likely reasons of accident cropping-up so as to boil down the likely causes to the minimum set.In case of the accident-occurrence,the alarming system should be able to give out the failure signals so as for the system operator to shutdown the well-head operation and put the wrongs right.Nevertheless,the above-said limitation can still probably affect the accident-cropping-up calculation for the cause-effect diagram is prone to a great deal of deviation.Secondly,we have made it possible to transform the fault tree into Bayesian network,through which the impact of the multi-states of the alarming system can be obtained and the uncertain impact of the level transmitter and level controller on the signal faults can be analyzed.In addition,the improved risk assessment of the wellhead separator can help to implement the effect of opinion of the experts on the basic event failure.In such a way of improvement,the accident-cropping-up probability can be worked out more accurately.Besides,since Bayesian network is in a position to predict the accident-occurring probability as well as do flashback inference(diagnosis) to work out the posterior probability of the basic events,it is also in a position to find out the most probable mode likely to cause the separator failure.In case any accidents occur,the most likely comprehensive key event-testing states could be followed:the low-level,ESD valve failure,alarming failure,operational errors,and so on,all of which are favorable for the users to develop their own necessary maintenance strategies.Thus,it can be seen that the separator's risk assessment can be done more effectively through the cross-fertilization between the two methods.