Study On Basic Oxygen Furnace Steelmaking Knowledge Acquisition And Endpint Control Model Based On Rough Set

BOF steelmaking is a complicated, multi-component, multi-phase and hightemperature physiochemical process with fast chemical reaction rate and manyinfluencing factors. The control core of the process is to control the endpointtemperature and carbon content of molten steel accurately. During the BOF steelmakingprocess, improving the accuracy of the endpoint control can reduce the tap-to-tap time,prolong the life of furnace lining, reduce the consumption of steel material, improve andstabilize the steel quality, reduce production costs and improve the BOF productivitysignificantly. Thereby, it is an important way to increase the competitiveness ofsteelmaking enterprises.For now, the endpoint control of BOF steelmaking seriously relies on the engineersand site operators’ experiential knowledge in the majority of domestic steelmakingplants. And the experiential knowledge is mainly acquired by summarizing theexperiences and statistically analyzing the operational data simply. Due to the unevenqualities of the situ technicians and operators and the complexity of the workingconditions for actual steelmaking, the stability and accuracy of the endpoint control byhuman experience and simple statistical analysis can not meet the productionrequirements. Currently, the knowledge discovery based on artificial intelligence hasbeen widely used in various areas, and has got very obviously good results in practicalapplication. However, this research is still in its infancy in the field of BOF steelmaking.The production process of BOF steelmaking which has a complex smelting mechanismis influenced by many factors. The data collected in the smelting process ismultivariable, nonlinear and much high noisy. Therefore, to explore a discoveringmethod which can adapt to the complex characteristics of BOF steelmaking process anddiscover the converter steelmaking production knowledge and setup a system model forknowledge discovering and endpoint controlling have an important significance intheory and application.For the characteristics of BOF steelmaking knowledge discovery and endpointcontrol, a BOF steelmaking knowledge discovery model and a rough set-neuralnetwork endpoint control model are established based on the rough set theory andneural network analysis. Using the proposed models, the BOF steelmaking productionknowledge can be acquired automatically and the BOF endpoint control method can be optimized. For the rough set-based BOF steelmaking knowledge discovery model, theproduction data can be preprocessed by cleaning, standardizing and straggling. Theimportant influencing factors of BOF steelmaking are served as the knowledgediscovery condition attribute and the molten steel carbon content and molten steeltemperature of BOF smelting endpoint are served as the knowledge discovery decisionattribute. Combined the association rules extract algorithm, the rough set theory methodis used to realize the BOF steelmaking knowledge discovery attribute reduction. TheBOF endpoint control model based on rough set-neural network combined withmethodological characteristics of rough set theory and neural network theory. Theminimum condition attribute set determined by rough set theory method is served as theinput of the neural network conditions and has a major impact on the decision attributeset. The model can effectively simplify the network structure and improve theadaptability, accuracy and computational efficiency of the neural network model. Then,using object-oriented graphical modeling technology, visualization technology and theoptional Microsoft’s Visual Basic6.0programming technology, a BOF steelmakingknowledge discovery and endpoint control system model is developed. It is much moreadaptable, flexible and reliable.The numerical experiments for knowledge discovery and endpoint control areperformed based on the Xinyu210t BOF steelmaking plant production process and datainformation. The results show that: BOF steelmaking knowledge discovery model basedon rough set method could achieve effective reduction of knowledge discoveryproperties. At the same time, the rule extraction algorithm which uses the minimum rulesupport and confidence can find knowledge rules hidden in the data and achieve theautomatic acquisition of BOF steelmaking knowledge. In BOF steelmaking knowledgediscovery process, with the increase of training data set, the number of knowledge rulesextracted increases correspondingly. However, the hit rate predicted by the knowledgerules does not necessarily improve and the operational efficiency of the system willreduce. Therefore, the data size of the training set and the representativeness of thetraining data have important impact on the rough set BOF steelmaking knowledgediscovery. With the increase of rule support and confidence, the predictive accuracy ofthe model knowledge rules become higher and the number of effective knowledge ruleswill reduce, but the coverage of knowledge should be significantly reduced withconstraint range narrow of knowledge conditions. Therefore, more reasonable parametervalues should be selected based on experiments and we should not just pursue of the larger rules support and trust. The input layer node propertie of rough set-nervenetwork model is that the input parameters of the end point prediction model aredetermined by analyzing the factors which have major impact on the endpoint of thesmelting steel endpoint of the carbon content and the temperature of molten steel basedon rough set model. Therefore, related to the conventional BOF neural networkendpoint prediction model, the number of nodes of input layer and hidden layer isreduced and the pertinence of the model network structure expression is increased. Also,the computational efficiency and accuracy of the model is improved effectively.Aiming at the complex BOF steelmaking process knowledge and endpoint controlproblem, the BOF steelmaking knowledge discovery model based on rough set and therough set-neural network model endpoint prediction model are established throughintroducing the rough set theory in this paper. And, model system based onobject-oriented method is developed. Method validation and practical application whichindicate the effectiveness of method and system are achieved by analyzing the specificBOF steelmaking production process data. Research results provide new methods andmeans for the BOF steelmaking knowledge discovery, end control optimization andproduction control.