Data-driven Blast Furnace Condition Analysis

With the development of science and technology,data generation and storage become more and more convenient.After years of development,blast furnace smelting as a pillar part of the steel industry has accumulated a large number of historical records.Stable furnace condition is a pre-requisite for blast furnace safety,high quality and low consumption of production.There have been many related studies on how to detect changes in the furnace conditions.However,the existing re-search contents have some shortcomings such as low detection accuracy,poor model portability and high data input requirement.In this thesis,after analyzing and correcting the collected real data,we developed methods to analyze and detect the blast furnace conditions.The contributions of this thesis include:First of all,in order to realize the local analysis and mining of blast furnace data,a three-level data transmission and storage framework for the host computer of the blast furnace system,the field server of the blast furnace and the laboratory data analysis platform was set up.In order to obtain the correct data for blast furnace condition analysis,this thesis analyzes the correlation between blast furnace parameters to determine the strength of the relationship between the parameters as a basis for the input of regression prediction model.In this thesis,support vector regression(SVR)is chosen as a:model.Taking cold wind flow as an example,the optimal combination of input parameters and training data duration is determined.Furthermore,in this chapter,we adopt a novel density based clustering method to analyze the working conditions of blast furnace.Through the analysis,the relationship among the data of normal furnace condition,the data of furnace condition adjusted by the blast furnace operator and abnormal furnace conditions has been confirmed.We further propose a method by utilizing the PCA dimensionality reduction and the two parameters of T2 and SPE to detect the furnace condition.