Modeling For Soft Measurement And Software Development Of Multi Quality Parameters Of Hot Metal Based On M-SVR

Blast furnace iron-making is the main method to product iron.The improvement of its technology and theory attracts many experts and scholars' attention both at home and abroad.The modeling and control of it is one of the frontiers problems in the field of metallurgy automation today.But due to the complex physical and chemical reaction inside the blast furnace,the coupling of parameters,nonlinear components,high temperature and dust or other harsh environment making the measurement doesn't work,the blast furnace iron making process is difficult to build model.In addition,the temperature of molten iron,Si content,S content,P content and other molten iron quality parameters is measured by off-line laboratory analysis which has a long delay,that quality information cannot be timely feedback affects the blast furnace system automatic control to be conducted.Therefore the suitable method for the optimization control of the quality parameters of hot metal is to build the soft sensor model of the quality parameters of the hot metal first.In the view of these questions,supported by the National Natural Science Foundation of major projects " High performance control of the basic theory and key technology research for the operation of large sized blast furnace",the thesis used the technique of data modeling,establish the multiple quality of molten iron parameters soft sensor model with the method of multidimensional support vector regression algorithm,and optimized the structure parameters of the model,designed and developed the software of soft sensor system to predict hot metal quality parameters online.The main work as follows:(1)The thesis introduced the quality parameters detection for hot metal,considering that the mechanism model is difficult to build,the thesis used the data model instead by the theory of statistical learning.According to the technological mechanism of the blast furnace,the thesis analyzed the dynamic coupling relationship of the parameters of the blast furnace.(2)The thesis conducted data preprocessing,trained sample,build the model of quality parameters of the hot metal with multi-output support vector regression(M-SVR).Different from conventional single output SVR or SVM,M-SVR can train multiple classification hyper planes to estimate multiple output variables at one time.By simulation experiments where M-SVR is compared with the BP neural network,the M-SVR model played better on estimating the sample data and when the mounts decreased.(3)To solve the M-SVR structure parameter selection problem,the thesis put forward the integrated evaluation criterion of multi output model,established the improved M-SVR model by comprehensive evaluation of multiple output and genetic parameters optimization.Comprehensive Evaluation of Multiple Output of model contains four aspects information:correlation coefficient,root mean square error,expectation and variance of error probability density function.Based on comprehensive evaluation index,the thesis used genetic algorithm to search the structure parameters of M-SVR model.The experiment proved that this method is useful to improve the M-SVR model,and the comprehensive evaluation of model accuracy can guide the optimization process to find optimal parameters.(4)The thesis maked Liugang 2#Blast Furnace as the research object of the application,designed and developed the online soft sensor software system for quality parameters of multi-molten iron.The system can run directly on the IPC,model input data from the database in the automatic is loaded and does pretreatment automatically;output real-time data can be displayed and loaded;the model can be re-training,the operation is stable;the interface is easy to use,all in all the system provides a good reference for the blast furnace operators.