地表水总有机碳含量紫外-可见光谱检测方法

总有机碳是以碳含量评价水质有机污染的指标，可以反映水体受污染程度。目前地表水总有机碳检测多采用现场取样后实验室分析检测方法，该方法存在费时费力、操作复杂、二次化学污染等缺点。紫外-可见光谱法具有环保、操作简便、可实时在线原位检测等优点，在地表水总有机碳检测中具有很好的应用前景。针对总有机碳检测问题，采用了一种基于自适应增强学习的区间偏最小二乘回归方法，该方法将总有机碳吸收光谱波段分为若干子区间，初始化训练样本权重, 依次在各子区间建立偏最小二乘回归模型，根据子区间模型预测误差率计算该子区间预测结果的权重系数，并更新下一子区间训练样本权重，最后将各子区间模型预测结果线性加权得到总有机碳的检测结果。实验配制总有机碳标准溶液浓度25～150 mg·L-1共43个样品，第一时间段采集35个总有机碳标准样品光谱分为训练集和测试集，建立并验证总有机碳检测算法模型。为评价算法模型鲁棒性，在另一时间段采集剩余的8个标准样品光谱进行反测验证。实验结果表明，采用基于自适应增强学习的区间偏最小二乘回归法建立的总有机碳定量模型具有较高的精度和鲁棒性，分组验证和反测验证的预测均方根误差分别为1.304和1.533 mg·L-1，均优于偏最小二乘回归和极限学习机方法。为进一步验证该方法的有效性，使用该建模方法预测生活污水的总有机碳含量。实际地表水样本取样于河北石家庄藁城污水处理厂排污口污水及河北先河公司园区的生活污水，经稀释后共获得50组地表水样本，采用SPXY方法分为训练集33组水样，测试集17组水样。在实际水样检测中，采用净信号分析方法进行光谱预处理，降低总有机碳与其他水质参数间的交叉干扰；分组验证预测均方根误差为3.26 mg·L-1，平均绝对值百分比误差为3.46%。综上所述，基于自适应增强学习的区间偏最小二乘回归方法，可以快速准确地对地表水中总有机碳进行检测，为在线水质总有机碳检测提供了方法支撑。

Detection of Total Organic Carbon in Surface Water Based on UV-Vis Spectroscopy

Total organic carbon is an index to evaluate the organic pollution of water quality based on carbon content, which can reflect the degree of water pollution. Currently, the detection of total organic carbon in surface water mostly adopts the laboratory analysis method after field sampling. This method has the disadvantages of being time-consuming and laborious, complex operation, secondary chemical pollution, etc. UV-Vis spectroscopy has the advantages of environmental protection, simple operation and real-time on-line in-situ detection. It has a good application prospect in detecting total organic carbon in surface water. The interval partial least squares regression method based on the adaboost algorithm (Ada-iPLSR) is adopted. In this method, the total organic carbon absorption spectrum band is divided into several sub-intervals. The training sample weight is initialized. The partial least squares regression model is established in each sub-interval in turn, the weight coefficient of the prediction result of the sub-interval is calculated according to the prediction error rate of the sub-interval model, and the training sample weight of the next sub-interval is updated. Finally, the prediction results of each sub-interval model are linearly weighted to obtain the detection results of total organic carbon.43 total organic carbon standard solution samples concentrations of 25～150 mg·L-1 were prepared in the experiment. 35 total organic carbon standard samples were collected in the first period, and the spectra were divided into training and test sets. The total organic carbon detection algorithm model was established and verified. In order to evaluate the robustness of the algorithm model, the spectra of the remaining 8 standard samples were collected in another period for test verification. The experimental results show that the total organic carbon quantitative model established by Ada-iPLSR has high accuracy and robustness. The root means square errors of group verification and test verification are 1.304 and 1.533 mg·L-1 respectively, which are better than partial least squares regression and Extreme Learning Machine methods. In order to further verify the effectiveness of this method, this modeling method is used to predict the total organic carbon content of domestic sewage. The actual surface water samples were taken from the sewage at the sewage outlet of Gaocheng sewage treatment plant in Shijiazhuang, Hebei and the domestic sewage in the park of Hebei Xianhe company. After dilution, 50 surface water samples were obtained. SPXY method was used to divide them into 33 water samples in the training set and 17 water samples in the test set. In the actual water sample detection, the net signal analysis method is used for spectral pretreatment to reduce the interference of other substances in surface water on the detection of total organic carbon. The root means square error of group verification prediction is 3.26 mg·L-1, and the average absolute value percentage error is 3.46%. To sum up, the Ada-iPLSR method can quickly and accurately detect the total organic carbon in surface water, providing a method support for the on-line detection of total organic carbon in water quality.