CN107561046A - A kind of sewage plant Tail water reuse method of real-time and system based on fluorescence water wave - Google Patents

Abstract

Provides a method and system for real-time monitoring of tailwater discharge from sewage plants based on fluorescence watermarks and deep mining of fluorescence data. A fluorescence spectrophotometer with a fluorescence probe measures the raw three-dimensional fluorescence data of water samples in real time. Parallel factor analysis (PARAFAC ) to extract the effective fluorescent components of water samples, so as to more accurately and comprehensively characterize the structural characteristics and intensity of dissolved organic matter in water, and use the SOM neural network to train the PARAFAC fluorescent component data, and observe the output sensitive neurons and the distribution Space, quickly classify water samples, and evaluate the status of tail water discharge (normal, abnormal) in real time. The analysis and monitoring method of the present invention reflects the discharge state of the tail water of the sewage plant in real time, has flexible control, high degree of automation, simple operation and low operation cost, and can automatically alarm in time and grasp the on-site situation when the tail water discharge is abnormal, so as to facilitate the sewage treatment plant as soon as possible. Process adjustment provides a basis to ensure the stable operation of the sewage plant.

Description

A method and system for real-time monitoring of tail water discharge from sewage plants based on fluorescent water patterns

technical field

The invention relates to a real-time monitoring system for tail water discharge of a sewage plant based on fluorescence water marks and deep mining of fluorescence data, which belongs to the analysis method in the fields of sewage plant operation management and water environment monitoring.

Background technique

With the growth of population and the development of national economy, the contradiction between supply and demand of water resources has become increasingly prominent. At present, except for a small part of the tail water discharged from the urban sewage treatment plant for reuse, most of it is directly discharged into the waters near the sewage plant. The tail water is the sewage after the biochemical treatment of the sewage treatment plant. The tail water has a large amount of water, contains nitrogen and phosphorus, toxic and harmful substances, pathogenic microorganisms, etc., and it will affect the water quality and ecology of surface water and groundwater after it is discharged into the environmental water body. . According to the calculation of about 70 billion tons of tail water discharged by China's sewage plants every year (Class A discharge standard), 3.4772 million tons of COD, 34,700 tons of total phosphorus, 1.0431 million tons of total nitrogen and 347,700 tons of ammonia nitrogen will be discharged. . Therefore, it is of great environmental significance and social value to carry out real-time monitoring of the tail water of the sewage plant to ensure that the tail water of the sewage plant is discharged up to the standard.

At present, sewage treatment plants usually carry out monitoring, testing and analysis according to national discharge standards, including ammonia nitrogen, total phosphorus, total nitrogen, COD, etc. Most sewage plants use daily manual monitoring and analysis, which is time-consuming and laborious, and cannot reflect changes in the concentration of pollutants in tail water in real time; in addition, indicators such as COD are only apparent indicators of organic matter in water, and cannot fully assess the structural characteristics and strength of organic matter. Express. Therefore, there is an urgent need for a monitoring and analysis method that can monitor the organic matter in the tail water in real time and reflect the structural characteristics and concentration of the pollutants in the water.

Dissolved organic matter (DOM) in water will emit emission light of specific wavelength under the irradiation of excitation light of specific wavelength, and different types have different positions, so three-dimensional fluorescence spectroscopy (EEMs) can be used to characterize the composition of DOM in water, and it is like a fingerprint The same corresponds to the type of water sample, which is called "fluorescent water mark". As an emerging water pollutant analysis technology, three-dimensional fluorescence spectroscopy has the advantages of high sensitivity and no damage to the sample structure. It is widely used in the qualitative and quantitative analysis of pollutants, the detection of rivers and lakes, and the traceability of pollutants. In addition, the three-dimensional fluorescence spectrometer equipped with an optical fiber probe can obtain fluorescence data online in real time, which is suitable for water treatment process control, online monitoring and early warning of environmental water bodies.

When identifying and judging fluorescent water marks in water bodies, the "peak-finding method" is usually used for qualitative analysis of fluorescent groups. However, the simple peak-finding method cannot analyze the overlapping fluorescent peaks, which increases the difficulty of qualitative analysis of fluorophores. At the same time, the three-dimensional fluorescence spectrum data belongs to massive high-order data. How to extract effective information from the massive high-order data and evaluate the water quality of water samples is still a problem that needs to be solved.

Contents of the invention

Aiming at the drawbacks of three-dimensional fluorescent peak overlap affecting the qualitative and quantitative analysis of water quality and the time-consuming and labor-intensive analysis methods of sewage plants, the present invention provides a real-time monitoring system and method for tail water discharge of sewage plants based on deep mining of fluorescent water marks and fluorescence data , a fluorescence spectrophotometer with a fluorescence probe measures the raw three-dimensional fluorescence data of water samples in real time, and uses parallel factor analysis (PARAFAC) to extract effective fluorescent components of water samples, so as to characterize the structural characteristics and characteristics of dissolved organic matter in water more accurately and comprehensively. Intensity, and use the SOM neural network to train the PARAFAC fluorescence component data, observe the output sensitive neurons and the distribution space, quickly classify the water samples, and evaluate the status of the tail water discharge in real time (normal, abnormal). The analysis and monitoring method of the invention can reflect the tail water discharge state of the sewage plant in real time, has high sensitivity, is quick and easy.

In order to achieve the above object, the present invention adopts the following technical solutions:

(1) The original three-dimensional fluorescence data of the tail water are acquired in real time. The fluorescence spectrophotometer is connected to the fiber optic probe through the fiber optic adapter, and the fiber optic probe is placed in the tail water discharge outlet of the sewage plant to obtain the original three-dimensional fluorescence spectrum data of the tail water in real time.

(2) Data preprocessing. The real-time 3D fluorescence spectrometer transmits the data to the computer processing system. Fluorescence data is preprocessed based on Matlab to eliminate Rayleigh and Raman scattering interference and improve spectral analysis efficiency.

(3) PARAFAC extraction. Based on the MatLab and DOMFluor software platforms, parallel factor analysis (PARAFAC) is used to analyze the three-dimensional fluorescence spectrum to obtain the number of effective fluorescent components, and extract the effective fluorescent components of PARAFAC to quickly identify water quality characteristics and realize the "mathematical separation" of fluorescence information.

(4) Data import and initial construction of SOM neural network. The effective fluorescent component intensity extracted by PARAFAC was imported into the SOM neural network as an input vector, and the initial construction of the SOM neural network was carried out.

(5) Discrimination and classification of water samples. Through the training of the SOM network, the neuron information corresponding to each water sample is obtained, and the tail water quality judgment and pattern recognition are carried out according to the location of the neurons to judge whether the tail water discharge is normal.

The tail water of the sewage plant is: the secondary effluent of the biochemical treatment of the urban sewage treatment plant, or the effluent after the tertiary treatment. The sewage is mainly urban domestic sewage, and a small amount of industrial wastewater is allowed. The effluent quality meets the requirements of the "Urban Sewage Discharge Standard" Level B and above emission standards, ammonia nitrogen ≤ 15mg/L, COD ≤ 60mg/L, total phosphorus 1mg/L, total nitrogen 20mg/L.

The technical features of the fluorescence spectrophotometer are: the fluorescence spectrophotometer has a real-time fluorescence data acquisition function, the optical fiber probe is placed in the tail water discharge port of the sewage plant, and is connected to the fluorescence spectrophotometer through an optical fiber adapter. The measurement parameters of the original three-dimensional fluorescence spectrum are: PMT voltage 800V, excitation and emission wavelength (Ex/Em) 220-450/280-550nm, wavelength error ± 1nm, slit width 5nm, scanning speed 12000nm/min, single scanning time less than 1min.

The preprocessing method of the raw data of the three-dimensional fluorescence spectrum is as follows: the data of the three-dimensional fluorescence spectrum is a high-order matrix and stored in a csv format. Based on Matlab, the original three-dimensional fluorescence spectrum data acquired in real time is preprocessed, and the fluorescence intensity in the fluorescence area (Ex, Ex±20nm) is set to zero to eliminate the influence of first-order and second-order Raman scattering. The data above Rayleigh scattering (in the range of 20nm) was set to zero to avoid the interference of Rayleigh scattering.

Described PARAFAC fluorescent component extraction method is: based on MatLab and DOMFluor software platform, adopt parallel factor analysis (PARAFAC) to extract the effective fluorescent component of PARAFAC in the three-dimensional fluorescence data in real time, and obtain the loading score (F _max of each fluorescent component) ), as the fluorescence intensity of the current component. The DOMFluor software package is freely available from www.models.life.ku.dk .

PARAFAC is a high-order data iteration algorithm that can decompose a three-dimensional fluorescence data matrix into three meaningful matrices: A, B, and C. See formula 1 for the principle and formula.

i=1,2,…,I; j=1,2,…,J; k=1,2,…,k

In the formula: x _ijk is the number of components; a _in , b _jn , and c _kn represent the elements of component matrices A, B, and C with sizes I×N, J×N, and K×N respectively, and e _ijk is Elements of the residual cube matrix.

At the initial stage of the first use of the fluorescence monitoring system, when the number of data collection exceeds 100 times, the PARAFAC fluorescence water pattern model is started to be constructed. PARAFAC models with different numbers of groups (2-10) are used to fit the fluorescence data. In order to avoid falling into the local optimal solution, the singular value decomposition (SVD) of the matrix is used to generate the initial value, and the PARAFAC model with a certain number of groups is constructed. The split-half analysis, residual and load analysis were used for verification, and the number N of PARAFAC fluorescent components that passed the verification was obtained. Finally, a PARAFAC fluorescent watermark model with N effective fluorescent components was constructed on this basis.

The SOM neural network data input and model construction include: taking the PARAFAC fluorescence component intensity (N) of the current water sample as an input layer and inputting it into the SOM neural network. At the initial stage of the first use of the fluorescence monitoring system, when the number of data collection exceeds 100, the computer system uses the PARAFAC fluorescence component intensities (100×N) of 100 water samples as the input layer to construct the SOM neural network model. Model construction includes four steps: data standardization, SOM network initialization, SOM network training, and SOM network clustering analysis. The model output layer contains a certain number of neurons, and each neuron is laterally connected to other neurons around it, arranged in the form of On the checkerboard plane, the distribution of neurons is divided into two types by the Davies-Bouldin clustering method: normal neurons and abnormal neurons, which respectively represent the normal discharge and abnormal discharge of tail water.

The SOM neural network water sample discrimination and classification process includes: for each data collection input water sample PARAFAC fluorescence intensity, through the training of the SOM network, observe the output sensitive neurons and the distribution space, and perform water sample analysis. Rapid classification, real-time assessment of the status of tail water discharge (normal, abnormal), and real-time response in the computer terminal system when the water sample is abnormal.

The maintenance points of the PARAFAC fluorescent component model and the SOM neural network model are as follows: when the monitoring system is in normal use, the fluorescent water pattern model based on PARAFAC and SOM is rebuilt once a month, and the original data of the water sample is the past six months/one year Three-dimensional fluorescence spectrum data of 500 water samples randomly selected within the range.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. It can more accurately and comprehensively characterize the structural characteristics and strength of dissolved organic matter in water in real time, and the monitoring is timely and accurate;

2. Flexible control, high degree of automation, simple operation and low operating cost;

3. Use the computing system software platform to realize automatic data analysis and platform display. When abnormal tail water discharge occurs, it can automatically alarm and grasp the on-site situation in time, provide a basis for the process adjustment of the sewage plant as soon as possible, and ensure the stable operation of the sewage plant.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Fig. 1 represents technical route and data analysis flow process of the present invention;

Figure 2 shows the original spectrum of a typical three-dimensional fluorescence spectrum of sewage treatment plant tail water;

Figure 3 shows the four effective fluorescent components extracted by PARAFAC;

Figure 4 shows the intensity of effective PARAFAC fluorescent components contained in different water samples;

Fig. 5 represents the U-matrix diagram and four component surface diagrams of the successfully trained SOM neural network;

Fig. 6 represents the monitoring water sample neuron position through SOM network neuron clustering distribution and training;

Fig. 7 shows the monitoring water sample neuron position trained by SOM network;

detailed description

Below, the tail water discharged from a sewage plant in the Yangtze River Delta region after secondary biochemical treatment is used as the monitoring object, and the present invention is described in further detail through specific implementation methods, and the feasibility and accuracy of the method of the present invention are verified. Apparently, the described embodiments are only some of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

Please refer to FIG. 1 . FIG. 1 is a system diagram of a real-time monitoring system for tail water discharge from a sewage plant based on fluorescent water patterns provided by an embodiment of the present invention.

(1) The original three-dimensional fluorescence data of the tail water are acquired in real time.

The monitoring object is the tail water discharged from a sewage plant in the Yangtze River Delta after secondary biochemical treatment. The wastewater from the sewage plant will meet the Class 1A discharge standard after biochemical treatment. A CaryEclipse fluorescence photometer (Agilent, USA) was used to measure the original three-dimensional fluorescence spectral data of the water body. The fluorescence spectrophotometer has the function of real-time fluorescence data acquisition. The adapter connects with the fluorescence spectrophotometer. The measurement parameters of the original three-dimensional fluorescence spectrum are: PMT voltage 800V, excitation and emission wavelength (Ex/Em) 220-450/280-550nm, wavelength error ± 1nm, slit width 5nm, scanning speed 12000nm/min, single scanning time less than 1min.

Figure 2 is a typical original spectrum of three-dimensional fluorescence spectrum of tail water. It can be found that the fluorescence characteristics of the tail water of the sewage plant and the receiving water body are very similar, and basically there are four fluorescence peaks in a relatively wide area, representing ultraviolet humic acids (peak A), ultraviolet humic acids (peak A), and ultraviolet humic acids (peak A). C), tyrosine-like proteins (peak B), and tryptophan-like protein substances (peak T), wherein the fluorescence intensity of tryptophan-like proteins (peak T) and ultraviolet humic acids (peak A) is dominant status. Generally, domestic sewage and water bodies with strong microbial activities often show strong protein-like fluorescence, and tryptophan-like proteins often represent soluble microbial metabolites. In addition, humic acid substances are often chemically stable, difficult to decompose, and difficult to be biologically utilized. Most of the organic matter in the sewage was removed by the AAO biological oxidation process, but some residual organic matter and microbial metabolites were discharged into the environmental water body along with the tail water.

(2) Data preprocessing.

The three-dimensional fluorescence spectrum data is a high-order matrix and stored in csv format. Based on Matlab, the original three-dimensional fluorescence spectrum data acquired in real time is preprocessed, and the fluorescence intensity in the fluorescence area (Ex, Ex±20nm) is set to zero to eliminate the influence of first-order and second-order Raman scattering. The data above Rayleigh scattering (in the range of 20nm) was set to zero to avoid the interference of Rayleigh scattering.

(3) PARAFAC model construction and effective PARAFAC component extraction.

At the initial stage of the first use of the fluorescence monitoring system, the PARAFAC fluorescence watermark model was constructed. Based on the MatLab and DOMFluor software platforms, parallel factor analysis (PARAFAC) was used in real time to extract PARAFAC effective fluorescent components in the three-dimensional fluorescence data, and the loading score (F _max ) of each fluorescent component was obtained as the fluorescence intensity of the current component. The DOMFluor software package is freely available from www.models.life.ku.dk .

PARAFAC models with 2, 3, 4, 5, 6, 7, and 8 components were used to fit the fluorescence data respectively. In order to avoid falling into the local optimal solution, the singular value decomposition (SVD) of the matrix was used to generate the initial value. The constructed PARAFAC model with a certain number of components was verified by split-half analysis, residual and load analysis, and the number of verified PARAFAC fluorescent components was 4. Finally, 4 effective PARAFAC models were constructed on this basis. PARAFAC fluorescent watermark model for fluorescent components. Figure 3 shows four effective PARAFAC fluorescent components, which are respectively recorded as humic acid substances (C1), humic acid substances (C2), tryptophan protein substances (C3), tyrosine protein substances ( C4). The half-crack verification analysis shows that the constructed PARAFAC model is robust enough, and the extracted four fluorescent components can reflect the fluorescence spectral structure characteristics of water body, as shown in Table 1.

Table 1 Fluorescence peak characteristics of four PARAFAC fluorescent components

Attachment: The value in brackets represents the position of the second peak

Fluorescence components C1 and C2 are double peaks, with a broad fluorescence area, and have absorption peaks in the ultraviolet and visible light regions, representing typical humic acid fluorescent substances. In addition, the fluorescence component C3 has the most The peak usually belongs to tryptophan protein; the peak position of fluorescent component C4 is ex/em 270/325nm, which is generally considered to belong to tyrosine protein.

Figure 4 shows the intensity and change of the effective fluorescent components of PARAFAC in the tail water and water samples downstream of the tail water obtained at different sampling times. Among them, the average intensities of the four fluorescent components in the typical tail water (the fifth sampling) are 107.1, 56.4, 114.6, and 48.3 R.U., and the fluorescent components C1 and C3 are the main fluorescent components, and their average intensity ratios are respectively 32.8% and 35.2%. During the whole monitoring period, the fluorescent components of C1 and C3 accounted for 65-75% of the fluorescence ratio, indicating that the DOM in the tail water of the sewage treatment plant is mainly composed of microbial metabolites and residual refractory humic acid substances.

(4) Construction of SOM neural network model.

Based on the extraction of fluorescent components by PARAFAC, the intensity of fluorescent components was input into the SOM neural network, and then the SOM neural network was initialized and trained to construct the SOM neural network model. The build process includes:

The first step is data standardization. A total of 32 PARAFAC fluorescence intensity data were standardized to ensure that the average value of the standardized data was 0 and the variance was 1, so as to avoid the influence of the difference in magnitude on the training results. After the data preparation is completed, the data samples are transformed into a standardized SOM data structure, which is the input data for training the network.

Step 2: SOM network initialization. The initialization includes the initialization of the weight vector and the initialization of the corresponding training parameters.

The third step: SOM network training. The training adopts the Gaussian function batch training method, which is divided into two stages: rough adjustment and fine adjustment. After learning and training, each type of fluorescent data input will have a specific mapping on the neural network, so that the mapped neurons of the fluorescent data will be finally obtained.

The fourth step: SOM network cluster analysis. The K-means algorithm is used to classify the weights of neurons in the competitive layer of the SOM network, and the number of clusters is automatically selected with the DBI value (Davies-Bouldin index). By calculating the Euclidean distance between neurons, the minimum Euclidean distance obtained is the central area of each type of neuron, and then combine the weights of multiple competitive layer neurons in each type as the representative feature vector set of each type.

After initializing and training the SOM neural network, the output layer is optimized to create a network with 13*11 neurons, including 143 neurons, and the final quantization error and final graph error are 0.245 and 0.001, respectively. Figure 5 shows the trained PARAFAC-SOM network. Figure 5-1 is a U-matrix matrix diagram. The small blue hexagons represent neurons, and the depth of the color represents the distance between the neurons. The lighter the color, the smaller the difference between the neurons and the closer the properties; the color Deeper indicates that the neurons are farther away, and the nature and gap are larger, which helps to determine the cluster boundaries.

Figure 5-2 to Figure 5-5 are component plane diagrams, each component plane diagram represents the prototype vector of the mapping layer, and units at the same position in each component plane have the same prototype vector. It can be seen that the concentration distributions of C1 and C2 components in the component surface diagram, and the concentration distributions of C3 and C4 components are similar, which indicates that C1 and C2 components may come from the same source, and C3 and C4 components also come from the same source . In addition, the concentrations of C1 and C2 fluorescent components are roughly inversely proportional to the concentrations of C3 and C4 components.

The Davies-Bouldin clustering discriminant method was used to determine that the number of categories was 2, and the results are shown in Figure 6. It can be found that all water sample data are generally divided into two categories: the first category - normal water samples, and the second category - abnormal water samples. The neurons tending to the lower right may cause abnormal water quality indicators due to some reasons, and they belong to abnormal water neurons.

(4) Discrimination and classification of water samples by SOM neural network.

In the daily monitoring process, for each water sample index, PARAFAC analysis is used to extract an effective fluorescent component of each water sample, which is input into the SOM neural network. Randomly input the fluorescence index data of two receiving water bodies (A-good water sample/No.4 water sample, B-abnormal water sample/No.6 water sample) into the neural network to obtain the corresponding See Figure 7 for neuron locations. It can be found that the water quality of the No. 4 water sample is better, and its data unit corresponds to the normal neuron (A), while the No. 4 water sample is classified into the second type of data, and it is judged that there is an abnormality in the water sample data at this time .

It can be found that the real-time monitoring system for tail water discharge from sewage plants based on fluorescent water patterns can realize real-time acquisition and analysis of fluorescence spectra of water bodies, quickly identify water quality characteristics, and timely distinguish and classify them, so as to formulate measures for various types of water quality. The changing work plan provides a basis for decision-making, realizes the supervision and early warning of the tail water of the sewage plant and the surrounding receiving water bodies, and helps managers to use the corresponding work plan in a timely manner.

Inspired by the above-mentioned embodiments according to the present invention, and through the above-mentioned description content, relevant workers can completely make various changes and modifications within the scope of not departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, but must be determined according to the scope of the claims.

Claims (8)

1. A real-time monitoring system for tail water discharge of sewage plants based on fluorescent water patterns, characterized in that: (1) The original three-dimensional fluorescence data of the tail water are acquired in real time. The fluorescence spectrophotometer is connected to the fiber optic probe through the fiber optic adapter, and the fiber optic probe is placed in the tail water discharge outlet of the sewage plant to obtain the original three-dimensional fluorescence spectrum data of the tail water in real time. (2) Data preprocessing. The real-time 3D fluorescence spectrometer transmits the data to the computer processing system. Fluorescence data is preprocessed based on Matlab to eliminate Rayleigh and Raman scattering interference and improve spectral analysis efficiency. (3) PARAFAC extraction. Based on the MatLab and DOMFluor software platforms, parallel factor analysis (PARAFAC) is used to analyze the three-dimensional fluorescence spectrum to obtain the number of effective fluorescent components, and extract the effective fluorescent components of PARAFAC to quickly identify water quality characteristics and realize the "mathematical separation" of fluorescence information. (4) Data import and initial construction of SOM neural network. The effective fluorescent component intensity extracted by PARAFAC was imported into the SOM neural network as an input vector, and the initial construction of the SOM neural network was carried out. (5) Discrimination and classification of water samples Through the training of the SOM network, the neuron information corresponding to each water sample is obtained, and the tail water quality judgment and pattern recognition are carried out according to the location of the neurons to judge whether the tail water discharge is normal. 2. Based on the method according to claim 1, it is characterized in that: the tail water is the secondary effluent of the biochemical treatment of the urban sewage treatment plant, or the effluent after the tertiary treatment, the sewage is mainly urban domestic sewage, and a small amount of industrial wastewater is allowed, and the effluent The water quality meets the first-class B discharge standard of the "Urban Sewage Discharge Standard", ammonia nitrogen ≤ 15mg/L, COD ≤ 60mg/L, total phosphorus 1mg/L, and total nitrogen 20mg/L. 3. The method according to claim 1, characterized in that: the fluorescence spectrophotometer has a real-time fluorescence data acquisition function, the optical fiber probe is placed in the tail water discharge port of the sewage plant, and is connected with the fluorescence spectrophotometer through an optical fiber adapter. The measurement parameters of the original three-dimensional fluorescence spectrum are: PMT voltage 800V, excitation and emission wavelength (Ex/Em) 220-450/280-550nm, wavelength error ± 1nm, slit width 5nm, scanning speed 12000nm/min, single scanning time less than 1min. 4. The method according to claim 1, characterized in that: the three-dimensional fluorescence spectrum data is a high-order matrix and stored in csv format. Based on Matlab, the original three-dimensional fluorescence spectrum data acquired in real time is preprocessed, and the fluorescence intensity in the fluorescence area (Ex, Ex±20nm) is set to zero to eliminate the influence of first-order and second-order Raman scattering. The data above Rayleigh scattering (in the range of 20nm) was set to zero to avoid the interference of Rayleigh scattering. 5. based on the method described in claim 1, it is characterized in that: based on MatLab and DOMFluor software platform, adopt parallel factor analysis (PARAFAC) to extract the PARAFAC effective fluorescent component in the three-dimensional fluorescence data in real time, and obtain the load of each fluorescent component Score (F _max ), as the fluorescence intensity of the current component. PARAFAC is a high-order data iteration algorithm that can decompose a three-dimensional fluorescence data matrix into three meaningful matrices: A, B, and C. See formula 1 for the principle and formula. <mrow><mtable><mtr><mtd><mrow><msub><mi>x</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi></mrow></msub><mo>=</mo><munderover><mo>&amp;Sigma;</mo><mrow><mi>n</mi><mo>=</mo><mn>1</mn></mrow><mi>N</mi></munderover><msub><mi>a</mi><mrow><mi>i</mi><mi>n</mi></mrow></msub><msub><mi>b</mi><mrow><mi>j</mi><mi>n</mi></mrow></msub><msub><mi>c</mi><mrow><mi>k</mi><mi>n</mi></mrow></msub><mo>+</mo><msub><mi>e</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi></mrow></msub></mrow></mtd></mtr><mtr><mtd><mrow><mi>i</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mn>...</mn><mo>,</mo><mi>I</mi><mo>;</mo><mi>j</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mn>...</mn><mo>,</mo><mi>J</mi><mo>;</mo><mi>k</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mn>...</mn><mo>,</mo><mi>k</mi></mrow></mtd></mtr></mtable><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>1</mn><mo>)</mo></mrow></mrow> In the formula: x _ijk is the number of components; a _in , b _jn , and c _kn represent the elements of component matrices A, B, and C with sizes I×N, J×N, and K×N respectively, and e _ijk is Elements of the residual cube matrix. At the initial stage of the first use of the fluorescence monitoring system, when the number of data collection exceeds 100 times, the PARAFAC fluorescence water pattern model is started to be constructed. PARAFAC models with different numbers of groups (2-10) are used to fit the fluorescence data. In order to avoid falling into the local optimal solution, the singular value decomposition (SVD) of the matrix is used to generate the initial value, and the PARAFAC model with a certain number of groups is constructed. The split-half analysis, residual and load analysis were used for verification, and the number N of PARAFAC fluorescent components that passed the verification was obtained. Finally, a PARAFAC fluorescent watermark model with N effective fluorescent components was constructed on this basis. 6. based on the method for claim 1, it is characterized in that: the PARAFAC fluorescent component intensity (N) of current water sample is used as input layer, input SOM neural network. At the initial stage of the first use of the fluorescence monitoring system, when the number of data collection exceeds 100, the computer system uses the PARAFAC fluorescence component intensities (100×N) of 100 water samples as the input layer to construct the SOM neural network model. Model construction includes four steps: data standardization, SOM network initialization, SOM network training, and SOM network clustering analysis. The model output layer contains a certain number of neurons, and each neuron is laterally connected to other neurons around it, arranged in the form of On the checkerboard plane, the distribution of neurons is divided into two types by the Davies-Bouldin clustering method: normal neurons and abnormal neurons, which respectively represent the normal discharge and abnormal discharge of tail water. 7. based on the method for claim 1, it is characterized in that: the water sample PARAFAC fluorescence intensity that comes in for each data acquisition, by the training of SOM network, observe the sensitive neuron of output and described distribution space, carry out water sample Rapid classification, real-time assessment of the status of tail water discharge (normal, abnormal), and real-time response in the computer terminal system when the water sample is abnormal. 8. Based on the method according to claim 1, it is characterized in that: when the monitoring system is in normal use, the fluorescent water pattern model based on PARAFAC and SOM is rebuilt once a month, and the original data of the water sample is random within the range of the past half a year/1 year The extracted 3D fluorescence spectrum data of 500 water samples.