Research And Application Development On BOF Oxygen Model Based On BP Neural Net

The melting composition and temperature of BOF steelmaking was mainlycontrolled by the given oxygen, it was extremely important to determine the oxygendemand for BOF steelmaking. This paper combined the mechanism method and theintelligent control method and built a BOF static model. In order to control the total oxygen demand in the static blowing process of Basicoxygen furnace (BOF), this paper analysed the chemical reaction principle of the BOFsteelmaking process, based on material balance and heat balance, the static BOF modelwas built, in addition, the BOF oxygen model of BP neural net was also developed. Bythe approaches of simulation and error analysed of model predictions, the results showedthat, the BOF oxygen model of BP neural net can predict and obtain the total oxygenamount efficiently with better calculation accuracy and adaptive capacity, finally, thecontrolling accuracy and target point hitting ratio of the oxygen model were improved.The main contents of this paper were as follows: Firstly, for the complex chemical reaction of the BOF steelmaking, the BOF staticmodel based on material balance and heat balance was built. According to the massbalance theory of chemical reactions, through calculating the various oxides content ofthe slag and predicting the amount of auxiliary materials and oxygen demand. Accordingto the heat balance theory, to judge whether adding exothermic compound or coolant inthe process of BOF smelting process is necessary. By calculating the difference betweenincome heat and outcome heat, deciding the amount of the initiate coolant or exothermiccompound, and recalculating the amount of auxiliary materials and oxygen demand. Then the BOF oxygen model of BP neural net was developed in order to predict theoxygen demand of BOF static model accurately. Through the chemical reactions amongthe analysed elements, the key effecting factors of the BOF total oxygen demand werecertified. The improved BP neural network algorithms with momentum and variable stepsize were applied to build model. According to the historical information of BOF, neural network were trained and model parameters and network structure were determined. The accuracy of prediction model were verified via the test data.Finally, through putting the same set of data into mechanistic model and neural network model, the each calculated oxygen prediction demand and the actual case are compared. According to the principle of variance analysis data, three sets of data were done mean square error analysis, observing the fluctuation of each data. The results showed that the oxygen demand prediction of BP neural network model was more accurate.