Deep mining of risk weaknesses for petrochemical processes based on alarm logs
There are highly dangerous factors in the complex petrochemical processes.The raw materials and products have the characteristics of flammable,explosive,toxic,or harmful.The major dangers would be easily caused by a slight carelessness.During the complex petrochemical processes,a great number of potential risk information is contained in the process alarm logs,which is conductive to reveal the root cause of danger incidents and prevent the occurrence of safety accidents.It is important to make full use of alarm logs for the complex petrochemical processes.Therefore,a deep mining method of risk weaknesses for petrochemical processes is proposed based on alarm logs in this work.Firstly,a word embedded technology-Word2Vec is introduced to pre-process the text-type alarm logs and make them to vectorial data,so the text-type alarm logs are converted and quantized.The Pearson correlation coefficient is further applied to analyze the relationship between these alarm logs and obtain the correlation matrix.Secondly,according to the theory of complex networks(CN),the correlation matrix should be transformed into a Boolean matrix,and then the risk character network could be established for complex petrochemical processes.Thirdly,the technique for order preference by similarity to an ideal solution(TOPSIS)is used to accurately assess the node importance of the established network model.This work involves three indicators:degree centrality,proximity centrality,and eigenvector centrality.Finally,the risk weaknesses of petrochemical processes can be deeply mined based on the priority of network node importance.A diesel hydrotreating unit is selected as the test case.Results show that the proposed method can accurately and effectively mine the process alarm logs with the alarm levels of"High High(HH)"and"High(HI)",which is consistent with the actual operating conditions of petrochemical processes.
risk weaknessesalarm logsnode importancecomplex networks(CN)technique for order preference by similarity to an ideal solution(TOPSIS)
NETL
NSTL
万方数据
