Research On Key Technologies Of Early Warning Of Hazardous Chemical Accidents Based On Data-driven Method

The hazardous chemicals industry which represented of petrochemical industry,are high-risk,with the characteristics of "high temperature,high pressure,toxic,harmful,inflammable and explosive".With the large,scale,and intensive development of hazardous chemicals industry,once accidents occur,disastrous economic losses,serious environmental disasters,and sever social impacts will be caused.The accuracy,instantaneity,and effectiveness of accident pre-warning have seriously affected the rapid development of China's hazardous chemicals industry.The painful accident experience warned us,a pre-warning and prevention system has to be established through the analysis and research of historical accidents,to prevent accidents,reduce losses,and achieve harmonious and scientific development of the hazardous chemicals industry.Aiming at the frequent occurrence of hazardous chemicals accidents,this paper studied a series of key technologies for the analysis and pre-warning of hazardous chemicals accidents,emphasized the importance of data in accident analysis and pre-warning,and proposed a complete data-driven analysis method,from the collection,processing and storage of hazardous chemicals accidents data,to the analysis,extraction and expression of hazardous chemicals accidents,and to the construction,feedback and verification of the pre-warning model.What's more,this method has been applied in engineering practice,a platform for pre-warning and control of hazardous chemicals accidents has been developed and constructed,which has been applied to emergency management offices and safety supervision institutions,and achieved excellent social and economic benefits.The key factors leading the occurrence of hazardous chemicals accidents,such as the unsafe behavior of human,the unsafe state of objects and the uncertain influence of environment.In view of the current situation of missing of hazardous chemical accidents data acquisition,an intelligent gateway for data acquisition of hazardous chemical accidents,which integrates intelligent identification of anomalous events,was developed independently to realize the data acquisition of the above three factors.Especially for the behavior characteristics of field personnel in chemical enterprises,video monitoring was used to extract features from five aspects,including color moments,edge histogram descriptors,color and edge orientation descriptors,color layout descriptors and scalable color descriptors.Local Binary Fitting(LBF)Model was adopted to identify anomalous events.Through the design of storage and management of main data of hazardous chemicals accidents,the data foundation is laid for subsequent large data analysis and mining.This paper analyzed the shortcomings of the traditional accident analysis method based on causality,proposed the concept of accident state vector,and vectored the accident state to solve the problem of quantitative expression of accidents.In order to determine the relationship between the dimension of the vectors,the form of the vectors,and the relationship between the vectors,the safety knowledge ontology model of the hazardous chemicals was constructed.The distributed multi-source heterogeneous data was integrated and analyzed,and the entity and relationship extraction were carried out to establish a knowledge map of hazardous chemicals accidents.Comprehensive,quantifiable vectors of hazardous chemicals accidents state were formed,through the deep exploration of the relationship between the factors leading to the occurrence of hazardous chemicals accidents.Compared with the traditional causality analysis method,all the factors that may cause the accident were retained,and the comprehensive expression of the accident was increased.Based on the accident state vector,an accident analysis and pre-warning model of hazardous chemicals was constructed by using the excellent learning ability and good generalization ability of support vector machine algorithm.Aiming at the problems that the traditional support vector machine incremental learning algorithm has lost support vector,an incremental learning algorithm for state vector distance was proposed.While retaining many accident related factors,the vector distance was calculated for each state vector,the state of the accident state vector was effectively judged,and the accuracy of the prediction result was high.This has positive significance for the prediction and prevention of hazardous chemicals industry accidents.Applying the above research results to engineering practice,the platform for pre-warning and prevention of hazardous chemicals accidents was constructed,and the“five-layer three-system” technical framework of the platform was proposed.At the data level,relying on the Security Monitoring Cloud project,the unified storage of data in different places could be realized.The national,provincial,municipal,district and county four-level database was constructed in the Security Monitoring Cloud.The data could be uploaded and stored step by step according to the requirements of different types of data acquisition frequency.At the application level,the national,provincial,municipal,district and county four-level hazardous chemicals accidents pre-warning,prevention and control systems were developed and deployed.Facing the regulatory needs of different levels of regulatory agencies,remote inspections and spot checks of accident pre-warning and real-time dynamic data from the critical supervision of hazardous chemicals enterprises were achieved,to meet the needs of governments at all levels for the safety production,supervision and inspection of hazardous chemicals enterprises.A platform for pre-warning,prevention and control of accidents in hazardous chemicals industry were designed,developed,constructed,and carried out industry demonstrations in many places,and achieved good application results.