Abstract: Taking into full account the complex process of a petrochemical plant， a risk evaluation model based on Bayesian networks for composite techniques was built， which integrates HAZOP， LOPA and Bow-tie analysis. Firstly， the system fault tree was prepared， which was mapped into corresponding Bayesian networks， and the GeNIe software was utilized for fault prediction and diagnosis with the two-way reasoning of Bayesian networks and for locating the Bayesian risk fault node which most likely leads to accidents. Secondly， the causes and consequences of deviation occurring in the node parameters were studied by combining HAZOP and LOPA， and the risk level was confirmed through the failure probability of identified protection layers. Lastly， Bow-Tie analysis was applied to the remaining higher risk events for identifying the key activities to continuously sustain IPL effectively and then preventing the serious technique safety accidents from happening. Moreover， this risk evaluation model was put into use in the delayed coking device of Liaohe Petrochemical Company.

Key words: Bayesian network, risk evaluation model, fault tree, LOPA, Bow-Tie analysis, delayed coking