基于LLE-SVR的水质COD紫外光谱检测方法研究

Study of UV Visible Spectrum-Based COD Detection Method for Water Quality Monitoring Based on LLE-SVR

化学需氧量(COD)是反映水体污染程度的重要指标之一。针对紫外可见光谱COD测量法数据波段多,易受干扰的问题,提出以局部线性嵌入法(LLE)结合支持向量机回归法(SVR)建立预测模型,来提高预测精度。首先,通过尝试预处理方法与模型分析方法(SVR和偏最小二乘回归法(PLSR))的不同组合来判断预测模型的效果,结果表明,"小波变换(WT)+SVR"效果较好。为了减少计算复杂度,提高运算效率,分别运用LLE和主成分分析算法(PCA)对数据降维,再分别结合SVR建立COD浓度预测模型。结果表明,利用"LLE+SVR"得到的COD浓度预测模型,其训练样本的均方误差为0.076030,测试样本均方误差为0.061477,分别小于"PCA+SVR"模型的0.216076和0.317303。这种方法使模型预测精度得到提高,为紫外可见光谱法检测水质COD浓度提供了一种可行的分析方法。

Chemical Oxygen Demand （COD） is one of the important indicators reflecting the degree of water pollution. In view of the problems of multispectral data and easy to be disturbed in UV visible spectrum-based COD measurement, a method combining Locally Linear Embedding （LLE） with Support Vector Regression （SVR） is proposed to build prediction model to improve prediction accuracy Firstly, different combinations of preprocessing methods and model analysis methods （SVR and Partial Least Square Regression （PLSR）） are tried to estimate the effect of the prediction model. The results show that ＂Wavelet Transform（WT）＋ SVM＂ is better in the model effect. Then, in order to reduce computational complexity and improve computational efficiency, LLE and Principal Component Analysis （PCA） are used respectively to reduce the dimensionality of the data and establish the COD concentration prediction models combined with SVR. The results show that the mean square errors of the training samples and the test samples are 0.076030 and 0.061477 in ＂LLE＋SVR＂ prediction model, which are less than 0.216076 and 0.317303 respectively in the ＂PCA ＋SVR＂ model. This method improves the prediction accuracy of the model and provides a feasible analysis method for the UV visible spectrum-based COD concentration determination in water quality monitoring .