Polycyclic aromatic hydrocarbons (PAHs) have teratogenic, carcinogenic and mutagenic properties, which seriously pollute the ecological environment and threaten the health of human beings and the growth of animals and plants. PAHs enter into the water environment through various ways, such as sewage discharge, atmospheric settlement and surface runoff. Because of their large variety and similar chemical properties, it is difficult to achieve rapid and accurate determination of the conventional methods, such as chemical titration and electrochemical methods. In order to realize the qualitative and quantitative analysis of the PAHs in complex system, this article is based on the three-dimensional fluorescence spectrum analysis, and combined with the ensemble empirical mode decomposition(EEMD) which is used to de-noise, and the self-weighted alternating trilinear decomposition(SWATLD) which is used to do two order correction, the ANA and NAP in water and pond water environment were analyzed and measured. The sample is prepared by a reasonable concentration, and the three-dimensional fluorescence spectrum of the sample is measured by the FS920 fluorescence spectrometer. The real spectral data can be obtained by eliminating the scattering of the spectral data by the blank deduction method. Then the EEMD is carried out to remove the noisy information and improve the signal to noise ratio. This method has the advantages of strong self-adaptive and simple parameter setting. The denoising parameters were compared with fast Fourier transform, wavelet filtering and empirical mode decomposition. Finally, using “mathematical separation” instead of “chemical separation”, the SWATLD algorithm is used to identify and predict the ANA and NAP in ultra pure water and pond water environment. The algorithm is not sensitive to the selection of the group fraction, and can be used to detect the multi component object simultaneously under the coexistence of unknown interferon. It has the “two-order advantage”, and the prediction results are compared with parallel factor analysis. The results show that the EEMD method makes the spectrum of ANA and NAP more regular and smooth, and the effective information is more prominent. The signal to noise ratio of the de-noised data is 16.845 2, the root mean square error is 11.136 6, the waveform similarity coefficient is 0.990 9, and this three indexes are better than the other de-noising methods, such as fast Fourier transform and empirical mode decomposition. It can achieve the denoising effect of wavelet filtering without setting a priori parameter. Using the SWATLD two order correction method, the decomposition spectra of ANA and NAP in the verified samples are basically consistent with the actual spectra, and the average predicted recovery rates are 96.4% and 104.2% respectively. The predicted mean square root errors are 0.105 and 0.092 μg·L-1 respectively. In the pool water samples with unknown interferon, the decomposition spectrum is still consistent with the actual spectrum. The average prediction recovery of ANA and NAP is 94.8% and 105.5%, respectively, and the root mean square root error is 0.067 and 0.169 μg·L-1 respectively. Compared with the parallel factor analysis, this two indexes have the advantages.