目前, 国内外铜矿品位分析多以化学分析法为主, 但由于化学分析法存在成本高、 时间长和污染物残留等缺点, 其相对配矿流程存在严重的滞后效应, 致使尾矿铜含量过高, 必然造成资源浪费。 开展斑岩型铜矿可见光-近红外光谱特征与建模研究是解决这一问题的有效途径。 以121个乌山斑岩型铜矿的化学分析与光谱测试数据为数据源, 分析了斑岩型铜矿可见光-近红外光谱特征, 以主成分分析法（PCA）、 局部线性嵌入算法(LLE)两种降维算法对原始光谱数据进行了处理, 所降维数分别为3维和5维, 同时利用遗传算法(GA)对原始光谱数据进行了波段选择, 共选取了467个最佳波段。 然后以BP神经网络为建模方法, 并分别以92个和29个斑岩型铜矿可见光-近红外光谱数据作为建模样本和测试样本, 建立了斑岩型铜矿可见光-近红外光谱的定量反演模型。 利用原始数据所建模型的品位反演平均绝对误差为0.104%, 利用主成分分析法、 局部线性嵌入算法、 遗传算法处理后的数据所建模型品位反演平均绝对误差分别为0.110%, 0.093%和0.045%, 由此可见, 利用主成分分析法处理后的数据所建模型品位反演精度较差, 利用局部线性嵌入算法处理后的数据所建模型品位反演精度略有提高, 而利用遗传算法处理后的数据所建模型品位反演精度有明显提高。 研究结果表明, 基于低品位斑岩型铜矿可见光-近红外光谱数据反演模型的品位分析具有一定的可行性, 为我国低品位斑岩型铜矿的品位快速检测提供了一种有效的手段。

At present, the analysis of copper grades at home and abroad is mainly based on chemical analysis. Due to the disadvantages of high cost, long time and residual pollutants, the chemical analysis method has a serious hysteresis effect on the relative ore blending process, resulting in that the copper content of the tailings is too high, which will inevitably lead to waste of resources. It is an effective way to solve this problem by conducting visible-near-infrared spectroscopy and modeling of porphyry copper deposits. Based on the chemical analysis and spectral test data of 121 Wushan porphyry copper deposits, the visible-near-infrared spectral characteristics of porphyry copper deposits are analyzed. The original spectral data is processed by principal component analysis (PCA) and local linear embedding algorithm (LLE). The reduced dimension is 3 and 5 dimensions respectively. At the same time, the genetic algorithm (GA) is used to select the band of the original spectral data. A total of 467 optimal bands are selected. Then, this paper takes the BP neural network as the modeling method, respectively uses visible-near-infrared spectroscopic data of 92 and 29 porphyry copper deposits as the modeling and testing samples, and establishes a quantitative inversion model of visible-near infrared spectroscopy for porphyry copper deposits. The average absolute error of the grade inversion model based on the original data is only 0.104%. The average absolute error of the grade model based on the model processed by the principal component analysis method, the locally linear embedding algorithm and the genetic algorithm is 0.110%, 0.093% and 0.045% respectively. It can be seen that the grade inversion accuracy of the model based on the data processed by principal component analysis method is poor, the accuracy of grade inversion model based on locally linear embedding algorithm is slightly improved, and the grade inversion accuracy of the model based on the data processed by the genetic algorithm is improved obviously. The research result shows that the grade analysis based on the inversion model of visible-near infrared spectroscopic data of low-grade porphyry copper deposits is feasible, providing an effective method for rapid grade detection of low-grade porphyry copper deposits in China.