水是一种有限的资源, 对农业、 工业乃至人类的生存都是必不可少的, 良好的水环境是可持续发展的重要保障。 对水质信息的科学监测, 是实现水资源优化配置与高效利用的基础。 联合国环境署(UNEP)与世界卫生组织(WHO)指出, 应当加强发展中国家的水质监测网络, 包括数据质量的保证和分析能力的提高。 光谱法作为一种新兴的水质分析方法, 相比传统的化学水质监测方法, 具有“响应速度快、 多参数同步、 绿色无污染”的特点。 传统单波长、 多波长的线性模型依赖于水体对特定波长的吸收特征, 不适用于多组分混合溶液且普适性较差。 因此, 提出了一种基于IERT的非线性全光谱定量分析算法, 建立适用于多组分混合溶液浓度预测模型, 达到利用全光谱信息来预测浓度信息的目的。 利用实验室配置的COD, BOD₅和TOC多组分混合溶液与NO₃-N、 浊度、 色度多组分混合溶液作为实验样本, 使用光谱仪采集样本的光谱曲线, 通过全光谱数据进行浓度预测实验, 结果显示, 对于COD, BOD₅和TOC多组分混合溶液, 本算法对于三种组分的决定系数(R²)分别为0.999 3, 0.991 4和0.999 3, 均方根误差(RMSE)分别为0.024 4, 0.057 7和0.000 4; 对于NO₃-N、 浊度、 色度多组分混合溶液, 决定系数(R²)分别为0.983 4, 0.868 4和0.981 0, 均方根误差(RMSE)分别为0.100 5, 0.326 4和0.120 2。 通过对比本算法与偏最小二乘(PLS)、 支持向量机回归(SVR)、 决策树(DT)、 极端随机树(ERT)对于同一组数据的实验结果, 表明: 在两组多组分混合溶液的实验中, 本算法对于其中各组分的决定系数(R²)均为最优, 相比于其他对比算法均方根误差(RMSE)均有大幅减少。 本算法可利用光谱信息对多组分混合溶液进行定量分析, 在计算时间相当的情况下, 可有效的提高浓度预测精度, 减少定量分析的均方根误差, 可为光谱法水质监测提供一种新的有效途径。

Water is a finite resource, essential for agriculture, industry and even human existence. A good water environment is an important guarantee for sustainable development. The scientific monitoring of water quality information is the basis for optimal allocation and efficient use of water resources. The United Nations Environment Program (UNEP) and the World Health Organization (WHO) pointed out that national water quality monitoring networks in developing countries should be strengthened, including improving analytical capabilities and data quality assurance. As an emerging water quality analysis method, spectral method has the characteristics of “fast response, synchronization of multiple parameters, environmental protection and pollution-free” compared with traditional chemical water quality monitoring methods. The traditional single-band, multi-band linear model, relies on the absorption characteristics of water at specific bands, and it cannot be used for multi-component mixed solutions and has poor universality. Therefore, this paper proposes a non-linear full-spectrum quantitative analysis algorithm based on IERT. The concentration prediction model suitable for multi-component mixed solution is established to use full spectrum information to predict concentration information. We use the COD, BOD₅, TOC multi-component mixed solution and NO₃-N, turbidity, chroma multi-component mixed solution configured in the laboratory as the experimental sample, use the spectrometer to collect the spectral curve of the sample, and conduct the concentration prediction experiment through the full spectrum data. The experimental results show that for COD, BOD₅, TOC multi-component mixed solutions, the determination coefficients (R²) of this algorithm for the three components are 0.999 3, 0.991 4 and 0.999 3. The root means square error (RMSE) is 0.024 4, 0.057 7 and 0.000 4. For the multi-component mixed solution of NO₃-N, turbidity, and colority, the coefficient of determination (R²) is 0.983 4, 0.868 4 and 0.981 0. The root means square error (RMSE) is 0.100 5, 0.326 4 and 0.120 2. By comparing the experimental results of this algorithm with partial least squares (PLS), support vector regression (SVR), decision tree (DT), and extreme random tree (ERT) for the same set of data, the results show that in the experiment of mixed solution, this algorithm is the best alternative to the coefficient of determination (R²) of each component.The root means square error (RMSE) has been greatly reduced compared with other comparison algorithms. This algorithm can use spectral information to analyze the multi-component mixed solution quantitatively. It can effectively improve the concentration prediction accuracy and reduce the root-mean-square error of the quantitative analysis in the case of equivalent calculation time. Moreover, this algorithm can provide a theoretical basis for spectral methods on water quality monitoring.