Comparative Analysis And Improvement Research Of The Converter End-point Control Model

Converter end-point controlling can affect product quality, production efficiency, and economic benefit directly, and have been an important index of measuring comprehensive engineering level of converter steelmaking. With the development of full continuous casting and production process, the constant improvement of hot metal pretreatment and refining method, the production rhythm of steelmaking speeds up quickly. The hitting ratio of end-point has great influence on the secondary refining and smooth operation of continuous casting, and makes converter end-pointing controlling particularly important in the whole steelmaking process. So researches of improving the hitting ration of end-point controlling have important significance in theory and application.The thesis firstly analyzes the current situation of converter end-point controlling technology, combing with the application of artificial neural network in the converter end-point controlling and comparing the improved converter end-point controlling model based on the hybrid BP-GA algorithm and the quasi-dynamic controlling model of converter and points out the deficiency of the model. As to low hitting ratio of two artificial neural networks and pure data modeling, the end-point forecasting model of BP neural network based on the time is established in the later steelmaking process on the basis of the steelmaking mechanism, referencing the actual statistical production law, comparing with detecting the liquid steel components and temperatures by the sub lance in the later converter smelting. In the treatment of data, the model combined with the steelmaking mechanism, added the statistical production law, in the treatment of the lance position introduced the oxygen utilization ration as the inputting parameters which is the quadratic function of the lance position, quantized the change of lance position in the converter steelmaking in order to improve the end-point hitting ration of the model. In the algorithm, this paper studies on the BP neural network, and improves the BP neural network in accordance with the neural network's local optimum and over fitting, adopts the method of introducing the cross-validation sample and improves the initialization of weight and activation function. In the construction of the model, time affecting the converter smelting is considered in the BP neural network end-point control model.According to historical data of Stb32 of 1# converter in the Panzhihua Iron and Steel plant, the end-point controlling model of BP neural network based on the time is used to forecast timely. The forecasting results showed that: at accuracy of |ΔT|≤15℃for temperature and |ΔC|≤0.03% for carbon content at the first turning down, the average prediction hit ration of temperature and carbon content of liquid steel are respectively 72.6% and 75.8%, both hit ratio are 58.1%, and having better prediction results.So, research has made some progress, the end-point controlling model of BP neural network based on the time provides a new research thought and means for some relative researches of converter end-point controlling.