为了实现对经过污水处理后的染整废水化学需氧量(COD)的在线监测, 同时为了克服国标标准法和快速分光光度法快速在线检测COD的诸多不足, 进行了紫外可见光谱结合多元分析方法检测混合染液COD的研究。 研究样本为自行配制的混合染液, 通过自主研发的紫外可见光纤传感检测系统检测混合染液的紫外可见光谱。 该检测系统可实现对较高浓度溶液在线原位检测, 而无取样、 稀释过程。 通过标准化(z-score)、 平滑(smoothing)、 一阶导数(1st derivation)等方法对原始光谱进行预处理并结合主成分回归(PCR)、 偏最小二乘回归法(PLS)等对混合染液建立吸光度-COD的回归预测模型, 并利用该回归模型预测预测集COD值。 实验结果表明, 针对混合染料溶液样本, 在PCR算法中, 采用3种光谱预处理方法, 其中标准化法得到的PCR回归模型预测精度最高, 决定系数R2=0961, 预测均方根误差(root mean square error of prediction, RMSEP)=218。 进一步研究发现, 使用标准化法预处理结合PLS算法得到PLS回归模型预测精度(R2=0974, RMSEP=196)高于PCR模型。 说明针对该样本, 利用标准化法进行光谱预处理后, 再建立PLS回归模型, 能够比较快速准确地进行混合染液COD预测。 同时也说明该自行研制的检测系统能够适用在线检测染液COD。 该研究可以为进一步实现在线原位检测染整废水COD以及其他水质参数奠定基础。

In order to realize the on-line measurement of chemical oxygen demand (COD) in dyeing wastewater after sewage treatment, and to overcome the disadvantages of the standard method and the rapid spectrophotometric method, a COD detection method based on UV-Vis spectroscopy with multivariate calibration was developed in this study. The mixtures of dyes were selected as the researching examples. A novel self-developed optical fibre sensor system was used to directly measure UV-Vis spectroscopy of dye solution. The system could realize on-line in-situ detection of the spectrum for higher concentration solution, without sampling and dilution process. In this study, both methods, principal component regression (PCR) and partial least squares (PLS), were used to develop regression models (Absorbance-COD) respectively to predict COD of training set and prediction set after the spectrum preprocessing including z-score, smoothing, 1st derivation, etc. With experiments using mixtures of dyes solution, the results show that PCR model with z-score preprocessing performed best in the three preprocessing methods (0961 for determination coefficient R2, 218 for root mean square error of prediction (RMSEP)). Further study identified that using z-score, PLS model (0974 for R2, 196 for RMSEP) achieved better results than PCR did. The results indicated that without the process of digestion, it was feasible to use UV-Vis combined with z-score preprocessing and PLS calibration for COD detection. The method could apply to the rapid and accurate determination of COD in mixtures dyes solution and proved to be successful in enhancing the prediction performance of COD and showed good potential to be applied in on-line water quality monitoring. The overall results showed that the self-developed system could be applied to on-line detection of COD in mixtures dyes solution. This study laid a foundation for further implementation of on-line in-situ detection of COD and other water quality parameters of dyeing wastewater.