基于前导事件和贝叶斯网络的储罐溢油动态风险评价

部分高风险危化品企业搬迁改造困难重重,为控制风险、保护周围人民生命和财产安全,有必要建立动态风险评价系统,对事故发生概率进行监控和预测。采用贝叶斯网络对事故发生概率进行定量分析。先在利用蝴蝶结模型辨识事故原因和后果的基础上,将其转化为贝叶斯网络模型;再导入"前导事件"信息和先验概率推导后验事故发生概率,量化分析事故发生随时间的变化概率;最后,以储罐溢流场景为例进行动态风险计算,结果表明,随化工装置生产时间和"前导事件"增长,元件失效概率和事故风险呈显著增长趋势。因此建议企业应重视"前导事件"并采取措施减少"前导事件",如优化检维修方案、及时更换关键部件、进行全面的事故调查等。

Dynamic risk analysis model for the oil-container overflow scenario based on the Bayesian network and precursor events

This paper is inclined to propose a dynamic risk analysis model for controlling the overflow scenario of the oil-container based on the Bayesian network (short for BN) and the precursor events. The reason why we have chosen such a topic is that the petrochemical plants themselves are not usually aware of fire risk fluctuation in the oil processing. The so-called precursor incident is used to refer to all kinds of abnormal events in the production process,such as the near-miss,event-alarm and actually taking place incidents,which can be reflected in the real-time safety situation of the processing system. Therefore,it is necessary,first of all,to identify and determine the accident causes,barriers, consequences and the mutual relations between the aforementioned events in the said scenarios indicated in the bow-tie model and their transformed results to BN according to a certain logic relationship between the nodes. In all such cases,BN can be used to present and work out the common cause-failure and the mutually interdependent or interrelated barriers to overcome the disadvantages of the Bow-tie model. As to the Beta distribution (β distribution),it is by nature the mathematic basis to update the posterior probability in Bayesian network. And,since the historical data or the experts-assessment and determination prior probability play the very important role in the risk prediction and forcast,the incident alarm database should also be taken as the source of the precursor incident data-information,which are inserted into the BN model in the final analysis. And,in this way,the storage tank overflow scenario can be chosen as a sample case to be applied to the method. Hance,the above research results tend to show that the increase of the precursor incidents may lead to the significant increase of the facility failure likelihood and the incident risk throughout the life time of the processing unit,particularly, the risk probability of vapor cloud explosion (VCE),which can be increased by approximately 6 times during the period of 4 years. Therefore,the above results remind us that enough attention should be paid by the industrial managers to the precursor incident signs or signals for the incidents and take proper actions to reduce the number of precursor incidents,for example,making efforts to improve the inspection and reinforce the maintenance plan and measures,as well as to replace the key facilities and investigate the incidents that have taken place systematically and thoroughly.