Development Of Artificial Intelligence Prediction Models For BOF Endpoint Manganese Content And Phosphorus Content

BOF steelmaking is a very important steelmaking method, which is in the dominant position. BOF steelmaking is a very complex, high-temperature, physical and chemic industry process. Just about the complexity of the BOF steelmaking process, it makes the prediction of the steelmaking endpoint very important. In the process of BOF steelmaking, the experiential jockey will analyze the Manganese content and Phosphorus content of the endpoint molten steel, and combine the sublance measuring information of temperature and carbon-content to decide whether to wait for the sampling analytic results or to fleetly tapping. This paper, which is based on the technical innovative project of National Committee on Economy and Trade "Development of New Generation Steelmaking Process Model Base and its Industry Application", introduces the research and development for prediction modeling technology of BOF steelmaking endpoint Manganese content and Phosphorus content. The research is based on the practical data of Baosteel's BOF steelmaking process, and it uses the artificial intelligence technology.Developed the prediction model for BOF endpoint Manganese content. According to the theory of metallurgy and statistics, the variable parameters of the model were chose, and the regression model and neural network model have been established. Though the regressive model is simple and visual, the simulated results are not ideal. Neural network model has strong abilities of nonlinear mapping and errors rectifying among artificial intelligence technology. Converter steelmaking is a complex nonlinear process. Better simulation results could be obtained if we adopt neutral network technology to establish a model on the base of choosing appropriate parameters. When the prediction error precision is |△Mn|≤0.025%, the shoot ratio of prediction is over 98%. The neural network prediction mode of endpoint Manganese has been adopted in local production.Developed the prediction model for BOF endpoint Phosphorus content. Using the reference of endpoint Manganese content prediction modeling, the two kinds of models of regression and neural network for endpoint Phosphorus content prediction was established. The neural network prediction model for endpoint Phosphorus content received the better result. When the prediction error precision is |△P|≤0.003%, the shoot ratio is about 83%. But the neural network model still cannot reach the request of local production. The more researches are needed. To cooperate with the local use of neural network prediction model, and resolve some practical problems, which cannot be settled by the existing emulational software, the C language source codes which is based on the Levenberg-Marquardt optimizing arithmetic of BP neural network has been developed. Basing on the use of dynamic memory, this program is easy to be maintained and can run on the local process computer.Because the neural network prediction model for endpoint Manganese content has reached the application request of industrial local, the microcomputer edition of neural network prediction model for endpoint Manganese content has been developed combining the practical local application environment. The microcomputer edition model has the same interface format with the application environment as the local process computer. While the local conditions allow, the microcomputer edition model can be transplanted to the process computer at once.