基于变量聚类分析的烧结矿还原性能预测模型

Prediction Model for Reducibility of Sintered Ore Based on Variable Cluster Analysis

对60组烧结矿还原性能指标(Reduction Index,RI)和化学成分的训练集进行变量聚类,通过最佳子集回归,建立烧结矿RI预测模型并进行异常值检验和误差分析,得出影响烧结矿RI最主要因素是SiO_(2)、MgO、TiO_(2)、R、FeO和CaO/TFe。模型R2达到99.88%,相对误差均在3%以下,精确度较高,能够实现烧结矿RI快速精准预测。

Variable clustering was carried out in terms of the 60 sets of Reduction Index(RI)of sintered ore and the training sets of chemical compositions.Based the optimal subset regression,the prediction model for RI of sintered ore was established and then abnormal values were tested and error values were analyzed.After that it was concluded that the most important factors influencing the RI of sintered ore were SiO_(2),MgO,TiO_(2),R,FeO and CaO/TFe.When the R2 of the model reached 99.88%,all the relative errors were below 3%with higher accuracy,which could enable the model to achieve the rapid and accurate prediction for RI of sintered ore.