Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Expert systems, knowledge engineering

Prediction Model Research and Application of Expert System for Converter Steelmaking

Author ZhouQiLong
Tutor ZhangHuiYi
School Anhui University of
Course Applied Computer Technology
Keywords BOF MST Cluster Regression Algorithm RBF Network Composite PredictionModel
CLC TP182
Type Master's thesis
Year 2012
Downloads 443
Quotes 0
Download Dissertation

Converter steelmaking is the main method of steelmaking in the world. In China someadvanced steel mills have been used the Technology of Dynamic Controlling on Converter.But for historical reasons, a lot of steel mills’ converter steelmaking is also relying on thetechnicians’ experience. But the technicians’ experience will be affected by many factors.Some variation will be occurred during the process and it will have an unsatisfactory result.For the financial and technical constraints, these mills can’t change to use converter dynamiccontrolling technology. To solve this problem, a static prediction model of expert system onconverter steelmaking was established to improve the hit rate, the converter output andquality. And this will also reduce the energy consumption and their costs.The prediction model is the core of the convert steelmaking expert system. Theoreticalmodel, incremental regression analysis model, BP neural network and RBF neural networkis the main model of the converter steelmaking prediction. The prediction of theoreticalmodel usually is unsatisfied because of a lot of assumptions and experience. Under a stableprocession the incremental regression analysis model has a desired result. Traditionalincremental regression analysis model chooses the last data as the sample and the accuracyof prediction is not high. Compared to BP neural network the RBF neural network needs lesstime to train. Currently the RBF neural network usually was trained by nearest neighborclustering algorithm or k-means clustering algorithm, etc. Using these clustering algorithms,the cluster number and the threshold is difficult to be determined. To improve theshortcoming of these three models, in this paper improved theoretical algorithm, regressionalgorithm based on sample-self-selection and the RBF neural network based on MST clusteralgorithm was proposed. At last the improved algorithms were combined by dynamicweighting method, and the new model was named composite prediction model. This modelwas used to predict the oxygen consumption and the amount of the materials to be added.The model was simulated by the practical data of a factory. The result shows that the oxygenconsumption prediction hit rate was100%if the deviation between predicted value and thepractical value of the oxygen consumption not above1.25%, and the limestone addition hitrate was93.7%if the deviation between the predicted value and the practical value of thelimestone addition is not above4%and the dolomite addition hit rate was87.5%if the deviation between the predicted value and the practical value of the dolomite addition is noabove10%.This thesis used the Object-oriented Technology and the UML modeling to analyzethe system, and used the C++programming language to realize it. The expert system guidesthe converter steelmaking of Shaoguan Iron and Steel Co online. In practical usage of theprediction to the34furnaces steelmaking, the hit rate was89.23%when the accuracy of theaimed end-point carbon content was±0.02%, and the hit rate was86.15%when the accuracyof the aimed end-point temperature was±20℃. Under the preceding condition thesimultaneous hitting rate of both temperature and carbon content was78.46%. It exceededthe anticipated results.

Related Dissertations
More Dissertations