基于XRF与RBF神经网络的锌浸出渣有价组分精准定量分析研究

锌冶炼浸出渣是湿法炼锌工艺产出的冶炼固废渣，占锌冶炼固废产出总量的75%以上，因含有Zn，Cu，Pb，Ag，Cd和As等多种有价金属元素，其资源化利用潜力巨大。然而由于其成分含量不稳定，检测精度不足等原因，导致关键元素的资源转化效率难以保证，因此对浸出渣关键资源组分的精准定量分析在锌冶炼行业绿色发展方面具有重大意义。该研究以Zn，Cu，Pb，Cd和As五种目标元素为分析对象，分别采用XRF工作曲线法和XRF结合RBF神经网络模型的方法对浸出渣目标元素定量分析，以相对误差、相对标准偏差作为两种方法的评价指标，对两种方法进行分析比较。首先采用标准添加法对工业现场采集的锌浸出渣配制浓度梯度样，并以此为标准化样品进行ICP-OES检测，随后将ICP-OES检测结果作为目标元素定量分析基准值，对浓度梯度样品进行X射线荧光光谱（XRF）检测，建立目标元素工作曲线，利用工作曲线对各目标元素进行定量分析。同时用XRF光谱数据构建输入矩阵、样品目标元素浓度构建输出矩阵，训练RBF神经网络来构建浸出渣中目标元素多元定标模型，并用此模型实现浸出渣样品目标元素预测。工作曲线法定量分析结果与ICP-OES基准值对比得到相对误差均值为8.5%，标准偏差均值为4.0%；RBF神经网络预测结果与ICP-OES基准值对比得到相对误差均值为0.18%，标准偏差均值为0.58%。结果表明，两种方法均能实现浸出渣样品目标元素的定量分析，但XRF结合RBF神经网络的方法能够对浸出渣样品进行精准定量分析和基体校正，分析结果准确性和精密度优于传统工作曲线分析方法。

Accurate Quantitative Analysis of Valuable Components in Zinc Leaching Residue Based on XRF and RBF Neural Network

Zinc smelting leaching slag is the solid smelting waste produced by the hydro-zinc smelting process, accounting for more than 75% of the total output of zinc smelting solid waste. Because it contains Zn, Cu, Pb, Ag, Cd, As and other valuable metals elements, it has great potential in resource utilization. However, due to its unstable composition content and insufficient detection accuracy, it is not easy to guarantee the resource conversion efficiency of key elements. Therefore, accurate quantitative analysis of the key resource components of the leaching residue is of great significance in the green development of zinc smelting. In this paper, five target elements of Zn, Cu, Pb, Cd, and As are the analysis objects，the method of XRF working curve and the method of XRF combined with RBF neural network model used to quantitatively analyze the target elements of the leaching residue. The relative error and Relative standard deviation are used as evaluation indicators of the two methods to compare the performance of the two methods. First, the concentration gradient samples of zinc leaching residue collected in the industrial field were prepared by standard addition method, used as standard sample and detected by ICP-OES. Then the detection result of ICP-OES is used as the reference value for the quantitative analysis of the target element, the concentration gradient sample is detected by X-ray fluorescence spectroscopy (XRF), to establish the working curve of target elements, the working curve is used to analyze each target element quantitatively. At the same time, the XRF spectrum data is used to construct the input matrix, the target element concentration of the sample is used to construct an output matrix, and the RBF neural network is trained to construct the multi-element calibration model of the target element in the leaching residue. This model is used to realize the target element prediction of the leaching residue sample. Compared with the ICP-OES reference value, the average relative error and standard deviation of the working curve method are 8.5% and 4.0%, respectively; Compared with the ICP-OES benchmark value, the average relative error and standard deviation of the RBF neural network are 0.18% and 0.58%, respectively. The results show that both methods can achieve the quantitative analysis of target elements of the leach residue samples, but XRF combined with RBF neural network can achieve the accurate quantitative analysis and matrix correction of the leach residue samples. The accuracy and precision of the analysis results are better than the traditional working curve analysis methods.