基于稀疏多核最小二乘支持向量机的浮选关键指标软测量

Soft sensor of key index for flotation process based on sparse multiple kernels least squares support vector machines

由于浮选性能受多种因素的制约,适宜的矿浆pH值是高效泡沫浮选的关键。针对pH值在线检测仪易受干扰、维护保养成本高等不足,结合泡沫浮选过程机理分析,以泡沫视频图像特征为辅助变量,将局部核函数和全局核函数加权组合,提高模型的学习和泛化能力,利用Schmidt正交化理论约简多核矩阵,减小计算量,建立基于稀疏多核最小二乘支持向量机的浮选矿浆pH值软测量模型。工业运行数据测试结果表明:所建模型具有预测精度高、反应迅速、稳定性好等优点,适于工业应用。

The pH value of pulp can directly influence the mineral froth flotation efficiency. Considering the poor stability of detectors and serious manual detection time-delay, a novel soft-sensor of pH is proposed combined with the analysis of flotation mechanism and convex combination of Gaussian and linear kernel function based on the sparse multiple kernels least squares support vector machines using image features as instrumental variable. Furthermore, the kernel matrices were reduced by Schmidt orthogonalization theory to lower the computational complexity. The experiment has verified the presented model performs high prediction accuracy, high efficiency and good stability.