Classification Modeling Of Pellet Quality Based On LSSVM

With the development of modern equipment manufacture, iron and steel industry has made great improvement, and much better material is in great demand day by day.As one of the artificial rich ores, iron ore pellet has become an indispensable metallurgical material because of its particular metallurgy characters. Generally, the qualities of the products are examined after they were produced. It is not convenient to adjust the process parameters for workers with the real-time operation status, thus, it inevitably affect the factory's benefits ultimately. Therefore, the prediction of the quality of the produced pellet has great significance.Based on the analysis the feature of the grate-kiln product technics, this paper proposes a method of pellet quality classification model, which combines rough set with least square support vector machine. By the means of this method, the relation between the operation parameters and product quality could be found.With respected to the model input, some rough set knowledge is used to simplify the input dimensions. Firstly, discrete the decision table by the way of using continuous attributes support discrete method, then reduce the discreted decision table with pawlak theory to discovery the relations between the operation parameters and product quality. Secondly, normalize the condition attributes to eliminate the affects because of different units and great absolute value diversity. Thirdly, with respected of the machine learning, we use least square support vector machine to train model, and lower the misclassification rate by choosing the kernel function and parameters value. By analyzing the result of the simulation, the classifier is proven to be efficient.