Research On MINLP Stochastic Modeling Method And Application Based On Data Analysis

There are many uncertainty factors in the production process of the process industry,such as the demand for products of the refinery production process in the petrochemical industry and the temperature change of the molten iron in the steel industry.Therefore,it's necessary to study stochastic modeling methods and their application in process industry production uncertainty problem.The research can not only help enterprises adapt to dynamic changes in production environment,but also increase production profits and ensure stable production operations.In this thesis,we focuses on the MINLP stochastic modeling method based on data analysis,and applies it to the refinery whole process production scheduling problem in the petrochemical industry and the molten iron temperature prediction data quality diagnosis in the steel industry.Firstly,we use the MINLP stochastic modeling method based on data analysis to generate the scenario tree to describe the uncertain factors in the process industrial production process.Secondly,with respect to the uncertain factors in the whole process and the hot direct supply process of the refinery production process,we formulate a MINLP model of the refinery entire process production scheduling problem based on data analysis,and design an improved OA algorithm based on Fix-Relax.Finally,we analyze data quality diagnosis characteristics in molten iron temperature prediction,combines scenario tree and support vector machine,design data quality diagnosis method based on data analysis,and further develop data quality in molten iron temperature prediction diagnostic system.The specific details are shown as the following:1)We adopt the MINLP stochastic modeling method based on data analysis to generate the scenario tree to describe the uncertain factors in the process industrial production process.The numerical experiment is carried out for the uncertainty of product demand in refinery.The experiment results show that the MINLP stochastic modeling method based on data analysis can obtain the scenario tree with the well quality and be solved eficiently.2)the uncertain factors and hot direct supply process in the entire process of refinery production,a MINLP model of the refinery entire process production scheduling problem based on data analysis is established and the improved OA algorithm based on Fix-Relax is designed.The numerical experiment on the uncertainty of product demand is carried out The experiment results show that considering the hot direct supply process can improve the hot direct supply ratio of materials,and the improved OA algorithm based on Fix-Relax is superior to the OA algorithm in terms of iteration numbers and solution time.3)Through the analysis on the characteristics of data quality diagnosis problems in molten iron temperature prediction,combines scenario tree and support vector machine,design data quality diagnosis method based on data analysis.The numerical experiments on the actual molten iron temperature prediction data of steel enterprises are carried out.The experiment results show that this method can improve the diagnostic accuracy.4)We design and developed a data quality diagnostic system for hot metal temperature prediction,embedding data quality diagnosis method based on data analysis to diagnose data quality.The developed system includes the functional modules like data management,data quality diagnosis,hot metal temperature prediction,and data result diagnosis,which can help the companies manage steel production process more accurately.