Prediction of silicon content in hot metal based on RS-LSSVM model

Abstract

A model for prediction of silicon content in hot metal is proposed based on two integrated algorithms: attribute reduction algorithm of rough sets (RS) and least square support vector machine (LSSVM). Rough sets theory is used to construct decision table, discrete attributes, rank the importance of attributes and reduce attributes based on weighting-coefficient cumulative estimation. The key factors are extracted as the input variables of LSSVM. The method can reduce the dimensions of the data and the complexity, and improve the efficiency of training and the accuracy of prediction. The data of the model are collected from No.6 Blast Furnace in Baotou Iron and Steel Group Co. of China. The results show that the LSSVM model based on RS attribute reduction has better prediction results than the model using other variables. The hit rate of silicon content in hot metal reaches 90% at the range of ± 0.1 % based on the proposed model, which can meet the requirement of practical production.