The Study Of BOF Steelmaking Control Model

Converter steelmaking is an important part of steel production process.The maintask is to remove the iron’s impurities by oxidation, and obtain the qualifiedcomposition and temperature of the molten steel. And blowing end-point control is animportant operation of converter steelmaking, and its level of automation technologyaffects molten mass at the converter blowing period directly. This paper is to solvethree problems that exist in50tons steel converter of Lianfeng Iron and SteelCompany. The problems are: the control level of the converter is not high; Theblowing process depends on manual operation; The hitting rate is very low. Bystudying the crafts of steel making and analyzing large amounts of data which arecollected from process of production, this paper establishes a converter steelmakingcontrol model, and design a NearestNeighbor algorithm which is based on rules andweight to get the best solution. We use the best solution’s blowing operationinformation to guide the new furnace. And this paper has solved two technicalpremise that are static batching and LD endpoint prediction.Converter steelmaking static batching model is built based on the principle ofmass balance and heat balance. The current slagging system of Lianfeng Iron andSteel Company is single slag method. When the range of slag alkalinity is decided, themodel uses the molten iron’s weight, composition and temperature to calculate themount of the next furnace’s slagging agent which consists mainly of lime amount,dolomite amount. When the error of slagging agent range during600kg, the model’saccuracy rate reaches73.4%and can help add ingredients. Besides, this model builtthe formula for the content of the molten steel such as manganese, phosphorus, sulfur,and established oxygen consumption formula for steel per ton.Converter steelmaking control model is established based on the principle ofcase-based reasoning. According to the different types of steel, we designed case baseto save the static and dynamic parameters of stoves which were detected one time andreached blowing end-point successfully. The parameters consist of iron information,scrap’s quantity, oxygen operation, the gun-bit operating parameters. We use theNearestNeighbor algorithm which is based on rules and weight to search cases whichare similar to or even consistent with the new case from the data base. If there weresome cases that are similar to the new case, then the new case refer the operation of the matched result under rules. Otherwise, the new case would be operated byworkers.BOF endpoint prediction model was established based on RBF neural networkwhich has a fast approximation speed, high convergence level, and othercharacteristics. The modeling steps consist of sample selection, network architecturedesign, data processing and learning algorithms selection. The sample was dividedinto training sample and prediction sample. We used MATLAB to achieve the model’ssimulation. The simulation results showed that when the final temperaturedeviation, and the carbon content of the deviation|C|≤0.03%, the hittingrate of molten steel’ carbon content and temperature reached65%.