基于最小二乘支持向量回归机的矿浆酸碱度鲁棒软测量

Robust Soft-Sensing of Slurry pH Using LSSVR for Mineral Flotation Process

针对酸碱度在线检测仪稳定性差、维护保养成本高等不足及人工检测严重滞后的问题,结合泡沫浮选工艺机理分析,以在线泡沫视频图像表观特征为辅助变量,采用最小二乘支持向量回归机(Least Squares Support Vector Regression,LSSVR)实现了泡沫浮选矿浆酸碱度的软测量.将不同特性的核函数凸组合以提高模型性能,并采用最近邻山峰聚类算法约简核矩阵,降低计算复杂度,利用偏最小二乘回归提高模型鲁棒性.工业运行数据仿真结果表明,建立的软测量模型能够连续在线检测矿浆的酸碱度,并获得了比标准LSSVR、加权LSSVR及多核LSSVR更高的预测精度,可满足工业要求.

Considering the poor stability of detectors and serious manual detection time-delay,a novel soft sensor was proposed based on least squares support vector regression（LSSVR） with sparsity using image features as instrumental variable.Firstly,multiple kernels were combined and the kernel matrix was reduced according to an improved minus cluster algorithm.Then the partial least squares regression was used to improve the robustness and precision of the soft sensor.The experiment verified the presented model which performs high precision and good reliability compared with standard LSSVR,weighted LSSVR and multiple-kernel LSSVR.