Application Research Of Water Pollution Traceability Based On Three-Dimensional Fluorescence Spectrogram Data Principal Component Characteristics And Competitive Neural Network Recognition Algorithm

So as to protect limited water resources and a good natural environment,reduce the harm caused by water pollution emergencies,and minimize the scope of accident impact,it is necessary to develop a scientific and rapid water pollution traceability method to quickly identify the source of pollution at the early stage of the accident.This study proposes a new method for water pollution traceability.This method is based on the main water-related enterprises in Anning City of Kunming City as the research object.Three-dimensional fluorescence spectrometer is used to scan water samples rapidly in the visible-ultraviolet spectrum band to extract the characteristic emission light spectrum characteristics of fluorescent organic compounds.Visual manual elimination method combined with growth curve deformation function is used as the data pretreatment method.Principal component analysis（PCA）was used as the feature extraction method of excitation-emission fluorescence matrix（EEM）,and competitive neural network was used as the classification algorithm of EEM principal component features to calculate the similarity of organic components in water samples,so as to achieve the purpose of traceability.The results show that:（1）From the perspective of three-dimensional fluorescence thermogram,the signal of fluorescence characteristic peak and background noise signal can be enhanced without difference after scattering is eliminated and then normalized;The signal of fluorescence characteristic peak can be enhanced differentially and the background noise signal can be suppressed by using sigmoid growth curve deformation function.（2）Compare method A:"Competitive neural network classification with original data directly"（total matching rate P=62.28%）with method B:Using the"PCA-competitive network"（P=83.77%）,it can be inferred that the scattered light will significantly affect the display of characteristic peak signal for the fluorescence data without visual manual elimination and normalization.In this case,the competitive neural network calculates the difference of the fluorescence intensity of the scattered light,resulting in positive error,which leads to high similarity of water samples and virtual high total matching rate of traceability.This matching rate is of no practical significance to the tracing research.（3）By comparing method A（P=62.28%）with the Tscore classification result of method B（P=83.77%）,the data distribution of the classification number obtained by the latter is more dispersed.Therefore,it can be inferred that PCA can also effectively extract the characteristic differences of the fluorescence intensity of scattered light.It is proved that PCA feature extraction can amplify fluorescence peak feature.（4）Method C:Using the"Visual manual elimination method-competitive neural network",although P did not increase in the end and there was no tractable application value（P=47.37%）,the classification numbers of different wastewater were no longer mostly the same.Therefore,it was inferred that what was compared in the classification was not the difference in the fluorescence intensity of scattered light,but the difference in the distribution of fluorescence intensity value of the characteristic peak of wastewater.（5）Method D:"Visual manual elimination method-PCA feature extraction-competitive network classification"is used for noise suppression and signal gain transformation without sigmoid growth curve deformation function,and the matching rate P_max is only 38.60%,which basically has no application value for water pollution tracing.The sigmoid growth curve deformation function Y=1/（1+exp（-k（20X-10）））is constructed to transform the spectral data that has been scatterized and normalized,which can obviously improve the matching rate of traceability.Meanwhile,k and classification items will affect the matching rate results.（6）Complete technical route E:"Visual manual culling method-sigmoid growth curve deformation function-PCA-competitive neural network"is based on visual manual culling method to eliminate scattering and perform linear normalization,sigmoid growth curve deformation function is used to suppress noise suppression and gain fluorescence signals,three-dimensional fluorescence features are extracted by PCA,and then classified by competitive neural network.Calculate the matching rate of accuracy evaluation target water and source water samples of source,（?） average of 64.35%,the maximum P_max reached 73.68%,the complete technical route is feasible and effective.