Research And Application Of Coke Quality Classification Based On Transfer Learning

Coke is the main raw material of blast furnace iron-making,and its quality stability affects the quality of blast furnace smelting to a great extent.Because it is impossible to measure the coke quality on line in the industrial field,it is necessary to use the method based on data to analyze the coke quality index.However,the training data characteristics of the coke quality model in the industrial field are easy to change dynamically and there are few available samples,which leads to the difficulty of online classification of coke quality.This thesis presents a coke quality classification method based on distributed adaptive transfer learning.According to the difference of distance between source domain and target domain in regenerative kernel Hilbert space theory,an adaptive transfer learning method based on dynamic selection of training samples and their weights was proposed to improve the utilization effect of source domain data.Aiming at the problem that the data distribution of source domain and target domain is similar,but the distance of target domain and source domain are dynamically changed after clustering,a new method of weight adaptive updating of weak classifier is proposed,so as to improve the precision of the integration result of weak classifier and establish a stable coke quality classification model.In order to verify the effectiveness of the method presented in this thesis,artificial data sets and actual industrial data are used for experiments,and compared with the classical machine learning algorithm.Experimental results show that the proposed method has higher accuracy and stronger stability when solving practical problems with dynamic changes of data characteristics.Combined with the algorithm proposed in this thesis,the coke quality classification system was developed and applied to the coke sulfur classification model of a steel mill to realize the production management of the enterprise.The long-term application results of this method in actual industrial field show that this method can accurately classify coke quality,which is of great significance for enterprises to reduce cost,improve resource utilization rate and realize structural upgrading.