Research On Quasi-dynamic Control System For End-point Of LD Converter Steelmaking Process

It is an important approach to improve the control for end-point of LD converter, shorten the smelt times of steelmaking, decrease the smelt cost, improve the quality of molten steel, achieve the standardized operation for LD converter process and production planning by forecasting the first turning down. The thesis aims at the integration of the BP neural network and genetic algorithm to build the better Quasi-dynamic model for forecasting the carbon in molten steel, temperature of liquid steel and time at first turning down.Firstly, advance in LD converter control technology and end–point control has been analyzed. The theory and arithmetic of BP neural network has also been introduced. Furthermore, the improved BP neural network training arithmetic and hybrid algorithm integrating BP and GA based on validation sample has been presented, with respect to the reasons of BP neural network which falls often into local minimization and over fit. The results show that the training arithmetic base on validation sample can improve the generalization ability of neural network. The hybrid algorithm based on BP and GA algorithm with generalization ability and rapid local researching ability is more efficient than BP algorithm on training neural network.In the paper, the features and present state of automation of LD converter steelmaking, the technological factors that effect on LD converter steelmaking process have been discussed in detail. Aiming at the present state of restricted detection equipment and lower percent of pass at first turning down for medium and small sized LD converter, the model based on the historical process data have built the relation between input parameters, including initial temperature, component and weight of molten iron, the blowing mode and oxygen supply mode, input modes of feed and accessories, and output parameters, including component, temperature of liquid steel and time at first turning down, for many types of steel in the converting process. According to initial condition and the blowing schedule, the carbon in molten steel, temperature of liquid steel and time at first turning down can be predicted in the different time of converting process. Furthermore, the offline prediction of temperature, carbon content of liquid steel and time at first turning down have been finished for different LD converters and steel type in the Vanadium-extracting and steelmaking Plant of PISCO. The model for forecasting the temperature and carbon in molten metal at first turning down has been tested by using two kinds of steel at accuracy of C≤0.02% for carbon content, T≤15℃for temperature and t≤1 min for the time at first turning down.It has shown that, for the steel Stb32 in 1# LD converter, the average hit ratio of prediction of both temperature and carbon content of liquid steel, and the time at first turning down are respectively 43.0% and 80.1%, and for the U75V type in 3# LD converter, the average hit ratio of prediction of both temperature and carbon content of liquid steel, and the time at first turning down are respectively 45.4% and 79.1%. At the same time, hit ratio of prediction of both temperature and carbon content of liquid steel with the model is higher than that made by operator in commercial practice.