基于互信息特征筛选偏最小二乘的激光诱导击穿光谱铁矿浆定量分析

目前检测矿浆品位相对准确的方法是传统化学分析,但周期长、有滞后性,无法实现在线检测。实验利用激光诱导击穿光谱(Laser induced breakdown spectroscopy,LIBS)在线、原位、快速等优点,分析了铁矿选矿过程尾矿浆中铁元素的品位值。由于LIBS采集到的光谱数据中存在大量对成分分析无用的冗余信息,进而增加了建模复杂程度,导致建立的模型精确度不够、泛化能力不强。因此,在偏最小二乘(PLS)模型基础上,提出了基于互信息特征筛选的偏最小二乘模型。实验结果表明,与传统的PLS模型相比,基于互信息特征筛选的偏最小二乘模型在分析精度上得到了明显改善,测试样品的决定系数R²从0.52提高到0.90,测试样本的平均绝对误差(MAEP)从2.87%下降到1.38%,总样本的平均绝对误差(MAE)从1.0%下降到0.60%。

Quantitative analysis of iron slurry based on laser induced breakdown spectroscopy combined with mutual information feature selection partial least squares method

At present, the relative accurate method for the detection of iron slurry grade is the traditional chemical analysis. However, this method has long testing period and lagging problem. Moreover, it cannot realize the online detection. Laser induced breakdown spectroscopy (LIBS) has some advantages such as on line, in situ and rapidness,etc. Therefore,the grade of iron element in iron slurry tailings from iron ore beneficiation was analyzed by LIBS. Since the spectral data collected by LIBS contained a lot of useless redundant information for composition analysis, the modeling complexity was increased, leading to insufficient accuracy and poor generalization ability of the established model. Therefore, refer to the partial least squares (PLS) model, the PLS based on mutual information feature selection was proposed. The experimental results showed that the analytical accuracy of PLS model based on mutual information feature selection was significantly improved compared to the conventional PLS model. The determination coefficient (R²) of sample increased from 0.52 to 0.90. The mean absolute error (MAEP) of testing sample decreased from 2.87% to 1.38%. The mean absolute error (MAE) of total samples decreased from 1.0% to 0.60%.