Study on Data Processing Algorithms of Three-dimensional Fluorescence Spectroscopy and Their Application in Detecting Contents of Fluorescencent Organic Matter

Author:Xu Zuo

Supervisor:wang yu tian


Degree Year:2017





Three-dimensional fluorescence spectroscopy is a means to measure the spectra and concentrations of the components of interest in the mixture.The basis of quantitative analysis of fluorescence spectroscopy is the linear relationship between fluorescence intensities and concentrations of the low concentration solution.The relationship of fluorescence intensities and concentrations cannot be approximated as a linear relationship due to inner effect existing in high concentration of the solution.It is difficult to determine the true concentration of the solution correctly using a single fluorescence intensity value.The quantitative relationship between fluorescence intensities and concentrations was analyzed by using mineral oils with low solubility in water and fulvic acid and humic acid with high solubility as the research object.The inconsistency of fluorescence intensities of varied components and mixture system concentration was used to enlarge the detected concentration range,which provide theoretical basis of instrument of a wide range concentration measurement of fulvic acid and humic acid.For three-dimensional fluorescence spectral data containing strong inner filter effect,the general methods are diluting the solution to the linear range or using fluorescence component absorption spectra for inner effect correction,which increasing the data processing steps.A method combining region summation of Quasi-Monte Carlo’s randomized points and extreme value of multivariate functions was proposed,which contained inner filter effect expression in the target equation and solved simultaneously to simplify the data processing steps.The method can reduce the interference of random error and improve the signal–to-noise ratio to a certain extent by random sampling and summation.The objective equation also can contain the spectral transform function to highlight the interest spectral information.The uniqueness of the minimum solution can be obtained by combining the multi-component fluorescence peaks information.Scattering will interfere with the measurement of fluorescence in the aqueous solution.The first-order Rayleigh scattering has the largest amplitude in the aqueous solution,which will obscure the fluorescence information coinciding with the scattering region.The main idea of available eliminating the first-order Rayleigh scattering method is avoiding the scattering region to carry out spectral analysis or to interpolate the scattering region using the data outside scattering region,which cannot get the real fluorescence peak information overlapping the scattering spectra.Base on the approximate symmetry of the first-order Rayleigh scattering spectrum and the physical mechanism of fluorescence emission,the symmetry subtraction method and the symmetry fitting method were proposed to eliminate the first-order Rayleigh scattering.The symmetry subtraction method is suitable for recorded dense emission points while the symmetry fitting method is suitable for the spare case.The methods achieve the fluorescence information extraction from the coincidence of the first-order Rayleigh scattering peak and fluorescence peak to make up for the lack of the existing first-order scattering methods.In addition to the scattering interference in the solution in the fluorescence data measured by the fluorescence instrument,there are also the systematic error of the instrument itself and the random error in the measurement.Instrument system error can be corrected by the instrument’s own calibration files,while random error is required to remove by the corresponding noise reduction algorithm.The effects of some common one-dimensional and two-dimensional noise reduction methods on the results of parallel factor analysis were compared.Five-point smoothing algorithm was used before parallel factor analysis of concentration estimated of interests in humic,pesticide,petroleum and their mixtures to compare the accuracy of estimated concentrations.The noise reduction algorithm can improve the results of parallel factor analysis of partial component estimations,while the improvement effect is not significant.For the long-term monitoring of the known fluorescent components in the fluorescence system,the data need to rapid process and the trend of component content trends is more concerned than the accuracy of component concentration estimation.A block multi-dimensional partial least squares method was proposed.Different methods were used to predict the interest components of the humic,pesticide,petroleum and their mixed system in environmental monitoring.The modeling time and the accuracy of the prediction were compared.The method improved the speed of operation compared to whole data modeled by multi-dimensional partial least squares analysis and improved the accuracy of the concentration estimations compared to insufficient spectral information in fewer data points to speed up the modeling speed.