| At present,the environmental protection of water resources has become a global issue and has drawn attention of all countries in the world,more and more scholars have carried out works related to water quality monitoring.UV-Vis spectroscopy relies on regression models and chemical method standards.In recent years,because of its simple measurement procedure and quick detection without secondary pollution,UV-Vis spectroscopy is widely used in measurement of nitrate concentration and other parameters in water.In this dissertation,combined with the dual-optical path active correction continuum spectral fine acquisition spectrometer designed by our laboratory,based on full spectrum fine analysis technique,the standardized data acquisition process,spectral data pre-processing and analysis methods in water quality parameter prediction are described in detail.The prediction performance of peak area modeling method,partial least squares(PLS)and backpropagation neural network(BPNN)are studied in depth.Optimization of the modeling algorithm for nitrate concentration prediction in various water bodies is built up with modeling interval optimization,modeling character wavelengths screening and data dimensionality reduction.The proposed algorithm and optimization model are verified in practical applications.It embodies the broad application prospect of UV-Vis absorption spectroscopy for predicting the nitrate concentration in water.The main research contents and innovation points of this dissertation are as follows:(1)A set of standard raw spectral data preprocessing procedures for diverse water bodies was proposed,including spectral information extraction and spectral data preprocessing of the samples.According to the characteristics of the dual-path active calibration continuous spectrum fine acquisition spectrometer,the spectral information of the nitrate in the solution was highlighted from the raw spectral data including the light source and the dark background noise.Combined with the spectral characteristics of the nitrate and the type of water,spectral data preprocessing methods for low turbidity water and high turbidity water were proposed.(2)iPLS-PA algorithm for predicting nitrate concentration in seawater was developed.Aiming at the problem that the single-wavelength method used in the practical application of nitrate concentration prediction in seawater can be easily interfered by other substances in the water sample,the modeling method of adopting characteristic peak area instead of characteristic peak position and embedding iPLS for absorbance interval selection was proposed.At the same time,the iPLS-PA algorithm was modeled and verified using the seawater sample of the typical sea area of Wheat Island,and the predicted result of cross-validated R~2 reached 0.9986.(3)LLE-BPNN algorithm for predicting nitrate concentration in high turbidity and high chromaticity water was proposed.Excessive turbidity and chromaticity in the water can shift the characteristic spectral position of the nitrate and affect its linear relationship between absorbance and concentration.In order to solve this problem,BPNN was used to predict the nitrate concentration in such waters to improve the prediction accuracy.Combined with LLE nonlinear dimensionality reduction,the BPNN modeling time was greatly shortened while maintaining local features between spectral data.The proposed LLE-BPNN algorithm was validated by using samples of different turbidity,chromaticity and nitrate concentration.The R~2 predicted by the cross-validation of nitrate concentration was raised from 0.7274 to 0.9556,and the RMSECV was reduced by 1.3259 to 0.6585,the modeling time was shortened from991.91s to 4.46s. |
PDF Full Text Request