【摘要】 软测量技术是解决精矿品位在线监测的一种有效技术手段。针对浮选生产过程大滞后、非线性、强耦合、难于建立机理模型的特点，采用XGBoost算法建立精矿品位软测量模型。利用蒙特卡洛异常数据诊断方法剔除建模样本中的异常数据，采用主成分分析方法对建模数据进行降维处理，化简模型结构，通过HQPSO算法对XGBoost模型的超参数进行优化，提高模型性能。仿真结果表明，精矿品位的预测误差在±2.5%之间，且预测结果能够正确反映精矿品位的变化趋势。更多还原

【Abstract】 Soft sensing technology is an effective means to realize the on-line monitoring of concentrate grade.Due to the characteristics of large lag,nonlinearity,and strong coupling in flotation production process,it is difficult to establish mechanism model. In this work,XGBoost algorithm is used to establish the soft sensing model of concentrate grade. Firstly,the abnormal data in the modeling samples are eliminated by Monte Carlo abnormal data detection method. Then the principal component analysis algorithm is used to reduce the dimension of the modeling data and simplify the model structure. Finally,the super parameters of the XGBoost model are optimized by HQPSO algorithm to improve the model performance. The simulation results show that the prediction error of concentrate grade is between ±2.5%,which can correctly reflect the variation trend of concentrate grade.更多还原