Algorithm Study On Three- Dimensional Excitation And Emission Matrix Fluorescence Spectroscopy Of Polycyclic Aromatic Hydrocarbons In Water

Polycyclic aromatic hydrocarbons (PAHs) have been recognized as one kind of typical toxic organic pollutants in water. It is of extreme importance to monitor the composition and concentration of polycyclic aromatic hydrocarbons in water for the security of water quality in China. Three-dimensional excitation emission matrix (3 DEEM) fluorescence spectroscopy with its high sensitivity and good selectivity is suitable for real-time monitoring organic pollutants in water. But those issues such as accurate recognition and quantitative analysis of pollutant components in the presence of interferents caused by coexistent components from the natural water are still not solved thoroughly. Therefore, algorithm study on data processing of three-dimensional excitation emission matrix has great theory significance and utility value for real-time monitoring polycyclic aromatic hydrocarbons.In this paper, starting with the model establishment of three-dimensional fluorescence spectra of polycyclic aromatic hydrocarbon pollutants in water, the component identification and quantitative analysis of polycyclic aromatic hydrocarbons with 3DEEM based on different algorithms methods are studied. The main research results are as follows:(1) Component identification methods of polycyclic aromatic hydrocarbons in water with 3DEEM are studied. Based on projection gradient, multiplicative iterative and alternating least square non-negative matrix factorization algorithm, component identification methods on phenanthrene, anthracene and pyrene in deionized water are studied. The results show that convergence speed of alternating least square is fast, and its similarities between resolved and standard spectra are high. Principal component analysis as the initial value can effectively improve similarities between resolved and standard spectra with the similarity coefficients all above 0.98. With the random initial values, three-dimensional fluorescence spectra of phenanthrene, pyrene, anthracene and fluoranthene in ionized water, reservoir water and river water are studied with sparse nonnegative matrix factorization on right hand factor algorithm. With all the similarity coefficients greater than 0.80, its results are better than that of alternating least square non-negative matrix factorization algorithm.(2) Quantitative analysis methods of polycyclic aromatic hydrocarbons in water with 3DEEM are studied. Three-way parallel factor analysis, alternating trilinear decomposition, multi-way partial least squares combined with residuals bilinearization and unfolded partial least squares combined with residual bilinearization are studied for concentration predictions of phenanthrene, pyrene, anthracene, fluorine, acenaphthene and fluoranthene in deionized water, reservoir water and river water. The results show that root mean square errors of phenanthrene, pyrene, anthracene and fluorene by three-dimensional parallel factor algorithm are 0.5 μg/L,0.2μg/L,0.6μg/L and 0.2 μg/L, relative errors are 8%,5%,14% and 4% respectively; root mean square errors of Acenaphthene and fluoranthene are 2.9 μg/L and 3.3 μg/L, relative errors 53% and 69% respectively. Compared with that by parallel factor algorithm, results by alternating trilinear decomposition show the results of phenanthrene are poorer, and that of anthracene, acenaphthene and fluoranthene are improved. The results of acenaphthene by multi-way partial least square algorithm combined with the residual bilinearization are improved substantially. Unfolded partial least squares combined with residual bilinearization shows advantage on concentration prediction of polycyclic aromatic hydrocarbons especially for components with low fluorescence intensity. Its root mean square errors of prediction for phenanthrene, pyrene, anthracene and fluorene by unfolded partial least squares combined with residual bilinearization are less than or equal to 0.4 μg/L, and the relative errors are less than or equal to 6%. The root mean square errors of prediction for acenaphthene and fluoranthene are, respectively, less than or equal to 1.0 μg/L and 1.7 μg/L, and the relative errors are less than or equal to 19% and 35%.(3) Quantitative analysis of polycyclic aromatic hydrocarbons in the complicated aquatic environment is studied. Four-way parallel factor analysis on fluorescence spectra of polycyclic aromatic hydrocarbons in water under different concentrations of humic acid is studied. Identification and quantifying of polycyclic aromatic hydrocarbons are realized, and the fluorescence quantum yields of polycyclic aromatic hydrocarbons under different concentrations of humic acid are obtained. The results show that similarity coefficients between resolved and standard spectra of four kinds of polycyclic aromatic hydrocarbons are greater than 0.98, and recovery rates of four components for concentration in the deionized water are between 90% and 110%. Furthermore, with fluorescence quantum yield, a correction method of three-way model is given for quantifying polycyclic aromatic hydrocarbons in water. The concentrations of polycyclic aromatic hydrocarbons in the presence of humic acid are predicted with correction method. The method is used to analyse four kinds of polycyclic aromatic hydrocarbons in river water adding humic acid and the recovery rates of concentration are between 82.0% and 102.0%.