Research And Application Of Blast Furnace Iron Hydrothermal Condition Prediction Model Based On Sparse Gaussian Process

As an important part in steel production,blast furnace ironmaking plays an important role in the national economy.In the process of ironmaking,iron and steel enterprises commonly use content of silicon to express the thermal state of blast furnace.In other words,its tendency could indicate the development of thermal state.Therefore,this thesis starts with the prediction of silicon content in molten iron and establishes accurate prediction models of blast furnace thermal state.As an important reference index of blast furnace smelting process,it is especially important to ensure the stability,smooth running and high yield of blast furnace.In this thesis,the research object is the blast furnace thermal state and the core content is the research on the prediction model of molten silicon content.Sparse Gaussian process as the main method to study point prediction and tendency prediction of silicon content separately.The main work of this thesis are as follows:(1)Analyze the data of blast furnace and study the methods of abnormal value replacement,noise removal and timing registration.Aim at the problems of outliers and spike noise in the continuous sampling data of blast furnace.This thesis proposes a method of processing outliers based on incremental threshold and an algorithm of signal reconstruction based on EMD to realize the accurate restoration of the continuous sampling data.Then,this thesis analyzes the lag between input parameters and silicon content of molten iron and proposes an input and output timing registration method based on the MIC,which provides accurate data for the subsequent application of data modeling technology.(2)Use data driven method to solve the problem of silicon content prediction in molten iron.The combination of input variable dimension and data sample dimension reduces the complexity of the model.In terms of the input variable dimension.Firstly,analyze the source of silicon in the blast furnace molten iron from process experience.The factors affecting the silicon content in the molten iron are determined as an alternative input to the model.Secondly,through the data correlation analysis,the candidate input variables with high coupling correlation are simplified and screened.In terms of the data sample dimension.A solution of’Sparsification’ is proposed to solve the problem that the size of the data model expands rapidly with the performance decreases.Based on approximate linear dependency(ALD)and subset of regression(SOR),the sparse and fast training of ferrosilicon content prediction model is realized.This algorithm not only preserves the performance of the Gaussian process,but also improves the training speed of the model while also making the model more streamlined.(3)Study tendency prediction of silicon content in molten iron from the perspective of regression and classification.Then the Tendency prediction method based on integrated learning is proposed to improve the accuracy.Through comparative analysis,we find that the classification modeling method has better adaptability in the prediction of the tendency of molten silicon content.In order to solve the problem of low precision of single SVM model,an integrated prediction model based on AdaBoost is proposed.The diversity improvement methods of ’algorithm diversity’ and ’model diversity’ are designed to maximize the accuracy of the integrated forecasting model.This thesis uses data of a factory to verify the above methods.When the relative error is±0.1%,the hit rate based on sparse Gaussian process model is 97%.It improves the hit rate compared to the traditional Gaussian model,and reduces the complexity of the model by nearly 90%.It proves that the method has good applicability in the face of high-dimensional and largescale blast furnace data.At the same time,the integrated prediction model for silicon content of molten iron presented in this thesis has a tendency hit rate of about 79%,which can meet the requirements of blast furnace production.