Abstract: Abnormal events involving hidden dangers, near misses and non-casualty accidents are the early warning and signs of serious accidents.Considering the high frequency of such incidents, the accident model was established to identify and correct unsafe factors in such incidents so as to effectively prevent serious accidents.According to the process features and accident characteristics of the tank area, the System Hazard Identification, Prediction and Prevention(SHIPP) model was improved, and the tank area accident model based on safety barriers was built.The cause-consequence relationship model was constructed by combining fault tree and event tree, which were mapped into Bayesian network to characterize uncertainty and conditional dependence.As for new evidence information, probability updating was implemented through Bayesian network update mechanism.Though the empirically learning on the field abnormal event data based on Bayesian theory, the uncertainty of prior probability can be reduced and the dynamic risk prediction of accidents in oil tank area can be implemented.A tank area case was analyzed and the results show that, the probability of material and energy release in the tank area is high, the overall safety performance tends to deteriorate, and the safety inspection and hidden danger investigation should be strengthened.The research results can provide scientific guidance and decision support for risk prediction and accident prevention of large oil tank area.