Research On The Carbon And Temperature Of Endpoint Forecast And Control For Bof Steelmaking

BOF steelmaking occupies an important position in the steel industrial world. The main task of its production process is to smelt molten steel with qualified composition and temperature. In the present technical conditions, the detection of the molten steel composition and temperature can not be continuous, many factors influence the molten steel composition and temperature, and boundary conditions of smelting process control changes frequently, all of these factors brought great difficulties to the accurate control of the molten steel composition and temperature in the smelting process. At present, the endpoint control of the majority of our small and medium-sized converter is still under the control of manpower. Workers operate according to the experience, which leads to the uncertainty of endpoint control. In practical process, since the endpoint temperature and carbon content can’t be exactly controlled,"rework" is a fairly common phenomenon which may cause huge economic losses to enterprise. Therefore, raising the level of end-point control of BOF steelmaking is important. Accurate forecasting of the end-point of the BOF steelmaking and appropriate control using the majorized technological parameter are the important prerequisite of organizing production rationally, improving molten steel quality and reducing the cost of smelting. Theoretical model and Neural Networks model have there own advantages in the field of end-point control of the BOF steelmaking. So research of the BOF steelmaking model based on the theoretical and Neural Networks is no doubt a direction that deserves more research.In this paper, materials rationing model is established based on heat balance theory and material balance theory. The parameters of the model can be set by the steel specialists, so the accuracy of the steelmaking model can be improved in the materials rationing stage. Molten iron, scrap steel, ministrant materials and other substances were added to converter. Molten steel, slag and other substances were produced after the smelting. The material added to the converter and the material produced from the converter will comply with the law of conservation of mass. The total heat before smelting and the total heat after smelting will comply with the law of energy conservation.The development of artificial intelligence technology provides a new thought for the BOF steelmaking endpoint control. In this thesis, we will study and improve the classical RBF neural network based on the K-means clustering algorithm. The RBF neural network based on the K-means clustering algorithm is the most widely used neural network now. It has many advantages such as simple structure, easy implementation, strong learning capacity, and so on. But it also has the disadvantages inherent in the K-means algorithm: the K value need to be set by the user and the initial cluster centers are uncertain. We will research on the impact of different K value to the learning capacity of the neural network, and try to improve the performance problem of RBF neural network caused by the inappropriate choice of the initial cluster centers. The improved neural network has stronger learning capacity and higher forecasting precision, in addition it avoid the problem of performance instability caused by the uncertainty of the initial cluster centers. Then, the improved RBF neural network will be used to predict the endpoint temperature and carbon contents.On the based of prediction model, we will try to combine the genetic algorithm with the RBF neural network to control the endpoint temperature and carbon contents. In the BOF steelmaking, the amount of oxygen and the amount of the scrap steel are the two main influence factors to the endpoint temperature and carbon contents, whose adjustment will bring important effect to the endpoint. In this thesis, the genetic algorithm will be used to find the optimalizing value of the amount of oxygen and the amount of the scrap steel. The optimization objective is to minimize the distance between the predict endpoint value and the enactment endpoint value. The Neural network forecast function serves as the fitness function of genetic algorithm. The results show that the technology can achieve effective control of the endpoint.The dissertation finally summarizes the work that has been done and forecasts the work that should be done in the next stage.