Study On Improved SVM Model For The Prediction Of Metallurgical Properties Of Pellet

Under the time-background of “Resolve capacity,Green production”,the steel industry is facing a serious challenge.As the essential raw materials for blast furnace charging,the metallurgical properties of pellet directly affect the quality of molten iron and normal operation of blast furnace.Therefore,before putting it into the furnace,the quality testing of pellet is necessary.However,the inspection is complicated and consumable.In this paper,from the perspective of “Pellet microstructure determines its metallurgy performance,and the metallurgy performance reflect its microstructure”,based on the mineralographic character,an improved Support Vector Machine(SVM)for predicting the pellet metallurgical properties is studied,which emphatically improves the efficiency of the pellet metallurgy performance evaluation.The main contents of the article are as follows:First,based on the principle of perspective SVM algorithm,three typical kernel functions of RBF kernel,polynomial kernel and Sigmoid kernel are analyzed emphatically.Based on the minimum error and the shortest running time of the algorithm,an improved strategy for the SVM kernel parameters selection and the combination of the kernel species is formulated.Second,based on the theory of genetic algorithm framework,the SVM kernel parameter as an individual solution space,the SVM prediction accuracy of single kernel function and the algorithm running time are defined as the forms of fitness function.An adaptive algorithm for kernel parameter selection is designed.An weighted complex SVM primary function of RBF kernel,polynomial kernel and Sigmoid kernel are built,the kernel species composite weight is individual solution space,the SVM prediction accuracy of composite kernel function and the algorithm running time are defined as the forms of fitness function,a self-adaptive composite algorithm of SVM kernel is designed.Finally,the SVM kernel parameters adaptive selection and the SVM kernel adaptive combination are organic coupling,and the design of the thesis target algorithm(improved SVM algorithm)is realized.Third,based on the original sample information for 200 groups,the pellet metallurgical performance index and mineral phase characteristics are studied by using image processing algorithm and PCA algorithm,and the elite sample set is built,in which the key characteristics of mineralogical phase and the grade label of pellet metallurgical property technical indices(Reduction Swelling Index,Reduction Index,Low Temperature Reduction Degradation)as input and output respectively.In this sample set,nine kinds of control algorithm are designed,they are the SVM with three kinds of single kernel function,the SVM with three kinds of kernel parameters adaptively selected and the SVM with kernel type adaptive composite and so on.Then,testing the superiority of the article target algorithm by algorithm control method.The research results show that the target algorithm is under the iteration threshold,the accuracy of RSI prediction is 100%,the accuracy of RI prediction is 98% and the accuracy of RDI prediction is 100%,which is significantly better than the nine kinds of control algorithm.The target algorithm is a highly efficient green algorithm to predict the pellet metallurgical properties,which can reduce the consumables of pellet quality testing and indirectly share the pressure of the capacity of the steel industry.It is worth popularizing in practice.