Study On The Prediction Of Burning Through Point According To In-time Exhaust Fume Analysis

Sinter ore is one of the important raw materials for blast furnace ironmaking,which is formed by sintering powdery iron-containing raw materials.The sintering production process has the characteristics of large hysteresis,strong coupling,and nonlinearity.In the sintering process,the burning through point is an important basis for the adjustment of the process parameters.However,because of the complicated on-site process,bad production conditions and many parameters involved,there is no relevant instrument to directly measure the specific position of the burning through point.The traditional prediction models of burning through point,based on parameters such as moisture,air volume,and layer thickness,are built using support vector regressor or neural networks.These models still present some shortcomings such as insufficient prediction accuracy,long time,and poor real-time performance.Aiming at the problems existing in the sintering process above,a burning through point prediction method based on quantitative analysis of exhaust fumes is proposed.By real-time detection and analysis of the concentration of SO₂,O₂and NO_x emitted during sintering,the random forests was used to establish the burning through point model to achieve accurate prediction of the burning through point.The main research work is as follows:1.Establish a continuous emission monitoring systemBased on the process flow of a 450m² sintering machine of a steel plant,a continuous emission monitoring system of sintering machine was established.An exhaust fumes analysis sampling probe was arranged at the air intake hole of the large flue dust collector.The concentration of SO₂,O₂ and NO_x,temperature,and pressure of the exhaust fumes during the sintering process was collected in real time by a gas analyzer and a exhaust fumes flow detector,which was used for subsequent analysis and modeling.2.Pretreatment of exhaust fumes dataThe data of 18210 groups were collected,which were denoised by wavelet analysis.The abnormal values of 7478 groups were eliminated by box plot,and data of 10732 groups were remained.The results show that the preprocessing process improves the data quality and provides the basis for the mining and extraction of data information.3.Establish a prediction model of burning through point based on exhaust fumes analysis and exhaust fumesThe methods of correlation heat map and parallel coordinate was used to evaluate the importance of the parameters.The concentration of SO₂,O₂,NO_x and gas temperature and pressure were determined as modeling data,and then the training samples and test sample data were divided.The parameters such as SO₂,O₂,NO_x concentration,temperature and pressure were input.The burning through point of the sintering is output.The algorithm of random forests was used to establish the prediction model of the burning through point based on the exhaust fumes.Through the mean square error analysis,the model parameters such as the number of decision trees and characteristic variables were optimized,and the prediction model of the burning through point was completed.The determined prediction model of burning through point was used to verify the test set data of 2683 groups.The results show that the prediction model of burning through point based on the algorithm of random forests has higher prediction accuracy and processing efficiency,96.95%and 16.77s,than the models established by support vector regressor or BP neural network based on same data.In addition,the prediction accuracy of the random forest model is increased by at least 0.0171 compared to the traditional model established using parameters such as moisture,layer thickness,and air volume.The prediction method of burning through point based on the exhaust fumes provides a new solution to improve the prediction accuracy of the burning through point and improve the generation efficiency of the sinter ore.