在分光光度法分析高浓度锌溶液中痕量钴离子浓度的过程中, 由于基体离子Zn(Ⅱ)与待测离子Co(Ⅱ)化学性质相近且基体离子浓度过高, 导致Co(Ⅱ)的光谱信号与Zn(Ⅱ)的光谱信号重叠, 大部分Co(Ⅱ)的光谱信号被Zn(Ⅱ)的光谱信号掩蔽, 在部分波长段溶液的吸光度与浓度呈现很强的非线性, 混合溶液加和性不佳, 因此无法通过全波段信息实现痕量Co(Ⅱ)浓度的检测。 该研究提出了间隔相关系数-偏最小二乘法对溶液的紫外可见光光谱进行波长点的筛选并建立吸光度-浓度模型。 首先设计实验获取锌钴混合溶液光谱图; 其后使用间隔分区的方法根据预测均方根误差这一指标挑选出Co(Ⅱ)的最优特征区间段, 以减少高浓度锌液的掩蔽作用并去除空白信息; 再在Co(Ⅱ)敏感区域使用相关系数法对吸光度矩阵进行波长点的精选, 所筛选出的点能最大程度保留Co(Ⅱ)灵敏度和线性度; 最后利用筛选出的波长建立PLS模型并计算出Co(Ⅱ)浓度。 将所提方法与全波段偏最小二乘, 分区偏最小二乘, 蒙特卡罗无信息变量消除偏最小二乘及竞争自适应加权偏最小二乘进行了分析比较, 结果表明所提出的方法所需波长数分别减少89.1%, 40%, 72.3%和81.7%, 模型精度分别提高64.6％, 33.3%, 38.7%和24.3%, 样本最大相对误差5.45%, 平均相对误差2.21%, 妥善的解决了高浓度锌液背景下痕量钴离子浓度检测问题。

In the process of detecting the concentration of trace cobalt ions in the high-concentration zinc solution by spectrophotometry, because chemical properties of Zn(Ⅱ) and Co(Ⅱ) are similar and the concentration of base ion is too high. Those facts lead to the overlap of the spectrum of Co(Ⅱ) and Zn(Ⅱ) which most of the spectral signals of Co (Ⅱ) are masked by the spectral signals of Zn(Ⅱ), strong nonlinearity of absorbance and concentration, and poor additivity in mixture solution in the partial wavelength. Thus it is difficult to determine the concentration of cobalt by the whole band information. In this paper, the interval-correlation coefficient-PLS is proposed to select the wavelength of the ultraviolet and visible spectrum of the solution and then establishes the absorbance-concentration model. Firstly, experiment was designed to obtain spectrum of Zn(Ⅱ) and Co(Ⅱ) mixture; Secondly , evaluation indicator-the predicted root mean square error was used to extract and optimize feature from full-spectrum data, which reduced masking effect of the high-concentration Zn(Ⅱ) and removed blank information.; Then the correlation coefficient method was adapted on the analysis of absorbance matrix in sensitive region of cobalt to select wavelength points detailed. The points selected finally could retain cobalt sensitivity and linearity in maximum degree; Finally, the partial least squares(PLS) model was established by the selected wavelength point to compute the concentration of Co(Ⅱ). The proposed method was compared with full band PLS, iPLS, MCUVE-PLS and CARS-PLS, and the result showed that with the proposed method the number of selected wavelength points respectively decreased 89.1%, 40%, 72.3% and 81.7%, and accuracy under this model respectively increased 64.6％, 33.3%, 38.7% and 24.3% compared with other methods in the background of the high concentration of zinc solution. The maximum relative error was only 5.45 % and the average relative error was 2.21%. The proposed method can properly resolve the problem of detecting trace cobalt ion concentration in high-concentrations zinc solution.