随着我国经济的高速发展, 地表水污染问题日趋严重, 实现地表水水质连续监测对于保障人类健康和保护环境至关重要。 亚硝酸盐氮浓度是水质评估的一项重要指标, 污染的水体对人畜及水产构成很大威胁。 利用紫外可见吸收光谱检测有机污染物已经成为水质检测的重要方法。 国内关于紫外可见光谱法检测亚硝酸盐氮的文献并不多, 一般采用对水样进行化学前处理然后再利用紫外分光光度计预测浓度。 这种方法不仅检测步骤繁琐, 耗时耗力, 对环境进一步造成污染, 而且无法实现实时连续检测。 无化学预处理的基于紫外可见吸收光谱法的亚硝酸盐氮浓度检测鲜有报道。 采用紫外可见光谱法针对地表水水质的无人值守自动连续监测方法开展基础研究。 配置了亚硝酸盐氮样本溶液, 设计了为期三天的实验, 每天分别测量所有样本的紫外可见光谱(记为D1组, D2组, D3组)。 首先将前两天的样本分别进行偏最小二乘回归法(PLSR)建模, 利用交互验证得到平均绝对相对误差(MAPE)分别为1.19%和1.85%, 这说明PLSR模型具有优良的预测精度。 其次, 为了验证PLSR模型在不同条件下的适应性, 取D1, D2的实验数据进行互相预测分析。 两天互相预测的MAPE分别为3.36%和4.51%, 低于5%, 说明PLSR模型具有良好的鲁棒性。 最后, 将D1, D2的全部样本用于建立最终的PLSR实测模型, D3的样本作为测试集。 测试集的MAPE为2.19%。 结果表明, 基于紫外可见光谱分析技术的PLSR算法对溶液中亚硝酸盐氮浓度的算法检测结果的MAPE均控制在5%以下, 优于同类文献的测量精度。 此外, PLSR模型建模过程简单, 运算时间短; 建好的模型结构简单更容易移植并固化到嵌入式系统中, 为后期开发设计便携式装置带来便利。 作为地表水亚硝酸盐氮浓度检测的基础研究, 可为日后地表水水质的精准快速检测提供指导。

With the rapid development of China’s economy, surface water pollution has become increasingly serious. Therefore, it is of vital importance to realize the continuous monitoring of surface water quality to ensure human health and protect the environment. Nitrite nitrogen concentration is an important index in water quality assessment. Polluted water poses a great threat to human, livestock and aquatic products. The detection of organic pollutants by UV-Visible absorption spectrum has become an important method for water quality detection. A few papers about the detection of nitrite nitrogen concentration in water by UV-Vis spectroscopy in China can be found. Most methods require chemical pretreatment of water samples and then use UV spectrophotometer to predict the nitrite nitrogen concentration. These methods are tedious, time-consuming and labor-consuming so thatthey can’t realize real-time continuous detection. Besides, they will cause further environmentpollution. The detection of nitrite nitrogen concentration without chemical pretreatment based on UV-Vis absorption spectrometry is rarely mentioned in the literature. Therefore, UV-Vis spectroscopy was adopted in this paper to carry out basic research on unattended automatic continuous monitoring of surface water quality. Nitrite nitrogen solution samples were preparedand a three-day experiment was designed to measure the ultraviolet and visible spectra of all samples (group D1, group D2, and group D3) respectively every day. Firstly, samples from group D1 and group D2 were modeled respectively by partial least squares regression method (PLSR). The mean absolute percentage error (MAPE) obtained by full cross validation was 1.19% and 1.85% respectively. The result shows that the PLSR model has good prediction accuracy. Secondly, in order to verify the adaptability of PLSR model under different measurement conditions, the experimental data of groupD1 and group D2 were used for mutual prediction analysis. The MAPE was respectively 3.36% and 4.51%, less than 5%, indicating that PLSR model has good robustness. Finally, all samples from groupD1 and group D2 were used for PLSR modeling, and samples from group D3 were used as the test set. The MAPE of the test set was 2.19%. The results show that the MAPE of the PLSR algorithm based on UV-Vis spectral analysis technique for detecting the nitrite nitrogen concentration in solution is controlled under 5%, better than similar reports. In addition, the modeling process of PLSR model is simple and the operation time is short. The model is simple in structure and easier to be transplanted and solidified into embedded systems, bringing convenience to the later development and design of portable devices. As the basic research of the detection of nitrite nitrogencon centration in surface water, this paper can provide guidance for the accurate and rapid detection of surface water quality in the future.