Multivariate Statistical Analysis Based Monitoring Of Blast Furnace Ironmaking Process

Blast furnace ironmaking process is the key process of iron and steel production,to ensure its continuous stability,low consumption and direct operation is the premise of iron and steel enterprises green low consumption production,improve product quality and reduce production costs.However,the blast furnace is a complicated and closed continuous production system.There are many variables of blast furnace,such as operation variables,state variables and index variables.How to use a lot of variables and their information to monitor the operation status and health level of BF to optimize operation system is the key to stable and low consumption operation of blast furnace.At present,the single variable threshold monitoring method is used more often in the blast furnace site.This method ignores the correlation between variables and coupling relation between different subsystems.Usually,the alarm occurs after the fault has reached a certain level,which causes inevitable loss and even endangers the life of the operator.In addition,the energy consumption of blast furnace ironmaking accounts for about 50%of the energy consumption of the iron and steel enterprises,and the fuel consumption accounts for 80%of the iron smelting energy consumption.Reducing the fuel consumption of blast furnace ironmaking is the key point of reducing the energy consumption of the blast furnace.Fuel ratio is a key index of blast furnace fuel consumption,based on the relationship between the operation variables and the data of the state variables of the blast furnace and the fuel ratio,the monitoring of the fuel ratio of the blast furnace can identify the key factors that affect the abnormal fluctuation of fuel ratio as early as possible.It is an important measure to reduce the energy consumption for the blast furnace.To solve the above problems,supported by the National Natural Science Fund Project"high performance large blast furnace operation control basic theory and key technology research"(61290323),2#BF in Guangxi as the research object,to carry out the research of measurement of blast furnace process based on multivariate statistical analysis,the specific work as follows:1)Analyzes the blast furnace ironmaking system,finds out the key parameters between the BF body and each subsystem,and sets up the monitoring data set,which contains 37 process variables,and then carries out the missing value and dimensionless pretreatment.Because non Gauss distribution and Gauss distribution data exist together in blast furnace process data,single multivariate statistical analysis method is difficult to describe the data distribution information of blast furnace running process completely,which makes the process monitoring appear false alarm and missing report phenomenon.To solve this problem,this paper considers the principal component analysis(Principal Component,Analysis,PC A),independent component analysis(Independent Component,Analysis,ICA)process monitoring algorithm characteristics and the complex situation of the existence of sub-gaussian distribution and super-gaussion distribution in the non Gauss distribution of process variables,a monitoring method of blast furnace process integrated with PCA-ICA is proposed,and the unified index and control limit of fault identification based on contribution diaGram are given.The algorithm is applied to the monitoring of the blast furnace operation process.The monitoring results show that the algorithm can achieve good results in the monitoring and abnormal identification of the whole process of the blast furnace.2)Aiming at the problem of fault identification in nonlinear complex industrial process system with less prior fault knowledge,a fault identification algorithm based on Kernel Partial Least Squares(KPLS)robust reconstruction error is proposed in this paper.Based on the kernel function technique,the algorithm firstly establishes the KPLS model of the quality variables and the feature space data under normal conditions.According to the relationship between the covariance matrix in the nonlinear mapping space and the Gram matrix in the kernel space,we can estimate the normal value of the process variables in the original space.The fault identification index and the corresponding control limit are calculated by the error of process variables,the variables beyond the control limit are the key factor that may cause the fuel ratio anomaly.At the same time,the algorithm considers the influence of the abnormal data in the new sampling data on the principal component of the feature space,and uses the iterative method to correct the score principal component,so as to improve the accuracy of the normal estimation of the process variables and the accuracy of fault identification.Finally,the simulation results of different fault types and multi fault variables,and TE process simulation results show the effectiveness of the algorithm.3)Analyzes the energy consumption in the process of blast furnace ironmaking,and uses the KPLS robust reconstruction error identification method combined with the Hatton control chart to monitor the fuel ratio of the blast furnace and identify the fault variables.The monitoring and identification results show that the algorithm can monitor the factors that have a potential impact on the fuel ratio in the process of blast furnace operation,and can effectively identify the abnormal source.