Method for inverting content of microcystic toxins in water by utilizing algae fluorescence spectrum

Abstract

The invention discloses a method for inverting the content of microcystic toxins in water by utilizing an algae fluorescence spectrum, which comprises the following steps of: firstly, measuring fluorescence spectrum data of N algae liquid samples with different concentrations, preprocessing, converting into a one-dimensional array Mp with the length of p, and further forming a matrix XN by using all the corresponding one-dimensional arrays; measuring the microcystin content of the N algae liquid samples to form a matrix YN; establishing a model Mridge based on the matrix XN and the matrix YN; and measuring fluorescence spectrum data of a sample to be detected, preprocessing the fluorescence spectrum data, converting the fluorescence spectrum data into a one-dimensional array Mpx with the length of p, further converting the one-dimensional array Mpx into a matrix Xp, substituting the matrix Xp into the established model Mridge, and automatically giving a predicted value of the microcystic toxins by the model. In the establishment process of the model Mridge, influence factors which influence the determination of the fluorescence spectrum and influence the growth of the algae can be added, so that the accuracy of the model is further improved. The method is more accurate, simpler, quicker and more convenient, facilitates on-site monitoring and has obvious practical value.

Description

Technical Field

[0001] The invention belongs to the technical field of water environment pollutant detection, and specifically relates to a method for inverting the content of microcystin (MC-LR) in water by using algae fluorescence spectrum. Technical Background

[0002] With the development of science and technology, more and more water bodies are seriously polluted, resulting in algal blooms and red tides in many rivers, lakes, and oceans. The most representative algal bloom is the cyanobacteria bloom. When the cyanobacteria explode and die, they will release a large amount of microcystins. Microcystins are the most common and most toxic toxins in freshwater areas. They are cyclic peptide hepatotoxins that are very harmful to animal and human health. The production of microcystins not only directly affects the safety of drinking water, but also causes an imbalance in the ecosystem, and enters the human body through enrichment in plants and animals, thereby causing human health problems, promoting tumorigenesis, poisoning, and even death. Therefore, real-time monitoring of microcystins in water bodies has always been an important topic for scholars to study. As an important indicator of drinking water, the limit value of microcystins shall not exceed 1.0μg / L.

[0003] The main methods currently used to monitor algal toxins are chemical, biological and immunoassay. Chemical method: It uses instrumental analysis methods such as high performance liquid chromatography (HPLC). The equipment is expensive, the pre-treatment steps are cumbersome and time-consuming, and it is difficult to achieve the purpose of real-time monitoring, especially for low-concentration MC-LR detection, which often results in large errors. Biological method: It is mainly measured by acute animal toxicity, usually by intravenous injection or gavage of mice, and is generally measured by the median lethal dose. This method is simple and crude and cannot be accurately quantified. Immunoassay: It uses an ELISA instrument to combine specific antigens and antibodies, and uses specific markers to make judgments. This method has high sensitivity, strong selectivity, stability and efficiency, but it is highly professional, time-consuming, and cannot obtain results immediately.

[0004] The patent with publication number CN 113029994 A discloses a method for inverting microcystin concentration based on the multi-source characteristic spectrum of extracellular organic matter, which establishes a relationship with microcystin based on chlorophyll a of the multi-source characteristic spectrum of extracellular organic matter of Microcystis aeruginosa. In practice, it is found that this method does not take into account the complexity of actual natural water bodies. For example, humus and water turbidity in natural water bodies will interfere with the fluorescence measurement of algae, causing a serious deviation between the predicted value and the true value, making it difficult to complete accurate quantification. Summary of the invention

[0005] The purpose of the present invention is to provide a method for inverting the content of microcystin in water using algae fluorescence spectrum, which solves the problem that the existing method does not take into account the complexity of actual natural water bodies, resulting in serious deviation between the predicted value and the true value and low accuracy.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for inverting the content of microcystin in water using algae fluorescence spectrum comprises the following steps:

[0008] (A1) Perform three-dimensional fluorescence scanning on N algae liquid samples of different concentrations to obtain the fluorescence spectrum data of the algae liquid samples, which are then converted into a one-dimensional array M of length p after preprocessing. p , the one-dimensional arrays corresponding to the N algae liquid samples are combined into a matrix X N , the matrix X N Each row of represents an independent variable of an algae liquid sample, and the matrix X N The size is N×p;

[0009] (A2) Determine the microcystin content of the N algae liquid samples in step (A1) to form a matrix Y N , matrix Y N The size is N×1;

[0010] (A3) Substitute the matrix X N With the matrix Y N Composition matrix M data , or the matrix X N After principal component analysis and dimensionality reduction, the matrix Y N Composition matrix M data , where Y N Located in the matrix M data The last column of the matrix M data Normalize and get the matrix M sdata , M sdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross-validation to automatically select the best regularization parameters to build a model. The established model is denoted as M ridge ;

[0011] (A4) According to the processing method of step (A1), the fluorescence spectrum data of the sample to be tested is measured, and the fluorescence spectrum data is converted into a one-dimensional array M of length p after preprocessing. px , and then the one-dimensional array M px Convert to matrix X p , and let X p X_Test, the predicted value of microcystin content in the algae liquid sample to be tested is Y_Pred, which is substituted into the model M established in step (A3). ridge In the example, the model automatically gives the predicted value Y_Pred.

[0012] In order to accurately invert the microcystin in algae water using fluorescence spectroscopy, it is also necessary to consider factors that affect the determination of fluorescence spectra and factors that affect algae growth (collectively referred to as influencing factors), such as light, water temperature, salinity, total alkalinity, turbidity, pH value, colored dissolved organic matter (CDOM), dissolved oxygen (DO), etc. To this end, an optimized method for inverting the content of microcystin LR in water using algae fluorescence spectroscopy includes the following steps:

[0013] (B1) Perform three-dimensional fluorescence scanning on N algae liquid samples of different concentrations to obtain the fluorescence spectrum data of the algae liquid samples, which are converted into a one-dimensional array M of length p after preprocessing. p , the one-dimensional arrays corresponding to the N algae liquid samples are combined into a matrix X N , the matrix X N Each row of represents an independent variable of an algae liquid sample, and the matrix X N The size is N×p. The influencing factors of N algae liquid samples are collected at the same time and expressed in matrix W N Indicates that the matrix W N The size is N×m, where m is the number of impact factors;

[0014] (B2) Determine the microcystin content of N algae samples with different concentrations to form a matrix Y N , the matrix size is N×1;

[0015] (B3) Substitute the matrix X N and the matrix W N Composition matrix M XW , the matrix M XW The size is N×(p+m), for the matrix M XW Perform principal component analysis to obtain the reduced dimension matrix M JXW , and then the matrix M JXW and the matrix Y N Composition matrix M JXWY , where Y N Located in the matrix M JXWY The last column of the matrix M JXWY After normalization, we get the matrix M Jsdata , M Jsdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross-validation to automatically select the best regularization parameters to build a model. The established model is denoted as M ridge ;

[0016] (B4) According to the processing method of step (B1), the fluorescence spectrum data of the sample to be tested is measured, and the fluorescence spectrum data is converted into a one-dimensional array M of length p after preprocessing. px , and determine the influencing factors of the samples to be tested, and convert the one-dimensional array Mpx The transformed matrix and the matrix of influencing factors are combined into the matrix X p , and let X p X_Test, the predicted value of microcystin LR is Y_Pred, and the model M established in step (B3) is substituted. ridge In the example, the model automatically gives the predicted value Y_Pred.

[0017] The algae solution samples in steps (A1) and (B1) and the samples to be tested in steps (A4) and (B4) include single algae solution samples and mixed algae solution samples, and steps (A1) and (A4), and steps (B1) and (B4) should correspond to each other. Specifically, the preparation of a single algae solution refers to taking a certain volume of algae species and culturing them to reach the required algae density of 10 7 cells / L or more, and a series of single algae solutions of different concentrations are obtained by diluting the algae solution of a certain algae genus of a certain algae phylum according to a certain dilution gradient; the preparation of mixed algae solution samples refers to taking a certain volume of algae of different algae genera of the same algae phylum that have reached the required algae density of 10 7 cells / L or more, by mixing or diluting algae in a certain proportion to obtain a series of mixed algae solutions of various concentrations; it may also refer to a series of mixed algae solutions of various concentrations obtained by mixing or diluting a certain volume of algae from different algae genera and phyla in a certain proportion to obtain a series of mixed algae solutions of various concentrations.

[0018] The parameters for the three-dimensional fluorescence scanning measurement in steps (A1) and (B1) are: excitation wavelength scanning range 351-660nm, excitation wavelength interval 5nm, excitation bandwidth 10nm, emission wavelength scanning range 451-800nm, emission bandwidth 10nm, scanning interval 1.0nm, and scanning speed 10000nm / min.

[0019] As an implementation method, the fluorescence spectrum data matrix with an excitation wavelength range of 351-660nm and an emission wavelength range of 451-800nm ​​obtained after three-dimensional fluorescence scanning is converted into a one-dimensional array M p .

[0020] As another preferred method, for the fluorescence spectrum data matrix obtained after three-dimensional fluorescence scanning, the data of the excitation wavelength 351-660nm and the emission wavelength 661-760nm are intercepted to form a matrix and converted into a one-dimensional array M y1 ; In addition, the data of excitation wavelength 541-620nm and emission wavelength 622-661nm are intercepted to form a matrix and converted into a one-dimensional array M y2 ; The one-dimensional array M y1 and M y2 Merge into a one-dimensional array M p .

[0021] In steps (A2) and (B2), the content of microcystin is determined according to GB / T 20466-2006 "Determination of Microcystin in Water".

[0022] The influencing factors in steps (B1) and (B4) include, but are not limited to, light, water temperature, salinity, total alkalinity, turbidity, pH value, colored soluble organic matter, and dissolved oxygen.

[0023] The pre-processing in steps (A1) and (B1) includes blank subtraction, Raman normalization, Rayleigh scattering and Raman scattering subtraction, data smoothing, and data matrix selection and optimization. Specifically, the following steps are included:

[0024] (101) subtracting the difference between the fluorescence spectrum data matrix of the algae liquid sample or the sample to be tested and the fluorescence spectrum data matrix of the ultrapure water;

[0025] (102) Based on the obtained Raman wavelength scanning data, the fluorescence data of ultrapure water emission wavelength range of 371-421 nm was intercepted and integrated to obtain the Raman integral of the ultrapure water Raman peak, which is denoted as Divide the fluorescence spectrum data matrix of the algae solution sample or the sample to be tested by the Raman integral after subtracting the blank The new data obtained are Raman unitized fluorescence spectrum data of the algae liquid sample or the sample to be tested;

[0026] (103) The influence of Rayleigh scattering is removed by setting the fluorescence intensity within the range of ±20-50 nm of the excitation light wavelength to zero; Raman scattering is the inelastic scattering of water molecules, and its influence is eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; the Delaunay triangle interpolation method is used to interpolate the data to further remove or reduce the influence of Rayleigh scattering or Raman scattering;

[0027] (104) Smoothing the data obtained in step (103) using a Gaussian kernel to reduce noise interference;

[0028] (105) Convert the data matrix obtained in step (104) into a one-dimensional array M p , or M px .

[0029] The method of the present invention utilizes algae fluorescence spectrum to invert the content of microcystin in water, obtains three-dimensional fluorescence data by scanning water containing algae, and takes water quality parameters affecting fluorescence measurement such as chlorophyll a, algae density, turbidity and colored dissolved organic matter as influencing factors, directly constructs a multivariate linear relationship between three-dimensional fluorescence spectrum data and microcystin through ridge regression algorithm, and constructs a model to calculate the content of microcystin-LR in water. The method is more accurate, simple, fast and convenient, is easy to implement on-site monitoring, and has obvious practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is the fluorescence spectrum of cyanobacteria.

[0031] Figure 2 This is the fluorescence spectrum of green algae.

[0032] Figure 3 This is the fluorescence spectrum of diatoms.

[0033] Figure 4 This is the fluorescence spectrum of dinoflagellate.

[0034] Figure 5 This is the fluorescence spectrum of cryptoalgae.

[0035] Figure 6 This is the fluorescence spectrum of sodium humate. DETAILED DESCRIPTION

[0036] In order to make the purpose and technical solution of the present invention more clear, the technical solution of the present invention is further described below in conjunction with embodiments, but the description of the present invention is not limited to the following.

[0037] Example

[0038] A method for inverting the content of microcystin LR in water using algae fluorescence spectrum

[0039] (1) Algae cultivation

[0040] (101) The dominant algae from five common phyla in my country’s natural water bodies were selected as the research subjects of the experiment. The dominant algae included Cyanobacteria, Chlorophyta, Bacillariophyta, Dinophyta, and Cryptophyta, and all were purchased from the Freshwater Algae Species Bank of the Chinese Academy of Sciences.

[0041] (102) The algae species purchased above were operated according to their instructions, the algae liquid in the test tube was shaken evenly, transferred to a sterile glass triangular flask in a clean bench, and placed in an intelligent constant temperature and humidity incubator for weak light culture for 2-3 days; the water temperature of the incubator was set to 25°C ± 1°C, the light conditions were 1000-4000 lux, and the light time was set to 12h day / 12h night; when the algae species grew well and the biomass increased significantly, they were transferred to the corresponding culture medium at a ratio of 1:5 again, and the triangular flask was shaken 2-3 times a day, and the position of the triangular flask was adjusted at any time to ensure uniform light, and the frequency of microscopic examination was 1-2 times a week. Among them, BG11 medium was used for Cyanobacteria and Chlorophyta, CSI medium was used for Bacillariophyta, 119 medium was used for Dinophyta, and AF6 medium was used for Cryptophyta. The culture media were all purchased from the Freshwater Algae Species Bank of the Chinese Academy of Sciences.

[0042] (103) When the algae growth reaches an algae density of 10 7When the algae density is greater than 100 cells / L or the fluorescence intensity of the algae solution measured by the fluorescence spectrometer reaches about 10000, step (2) is performed. The algae density is counted according to the standard HJ 1216-2021 "Determination of Phytoplankton in Water Quality - 0.1 mL Counting Frame - Microscope Counting Method".

[0043] (2) Preparation of algal liquid

[0044] (201) The preparation of algal liquid is divided into the preparation of single algal liquid and the preparation of mixed algal liquid.

[0045] The single algae liquid preparation is to take a certain volume of algae and culture it to reach the required algae density of 10 7 A series of single algae solutions of different concentrations obtained by diluting an algae solution of a certain algae species of a certain algae phylum with a concentration of more than 1 cells / L according to a certain dilution gradient.

[0046] The mixed algae solution preparation refers to taking a certain volume of algae of the same algae phylum but different algae genera that have reached the required algae density of 10 7 cells / L or more, mixed or diluted in a certain proportion to obtain a series of mixed algae solutions of various concentrations; or a series of mixed algae solutions of various concentrations obtained by mixing or diluting a certain volume of algae of different algae genera in a certain proportion;

[0047] (202) Take the above-mentioned certain volume to reach the required algae density 10 7 Cells / L or above various algae solutions were diluted to about 60%-80% of the fluorescence intensity of the initial concentration.

[0048] (3) Determination of various parameters

[0049] A certain volume of a single algae solution or a mixed algae solution of different concentrations prepared in step (2) is taken as a sample to complete the determination of parameters such as pH, conductivity, dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, colored dissolved organic matter (CDOM), and microcystins. Among them, the pH determination is carried out in accordance with the standard HJ 1147-2020 "Electrode Method for Determination of pH Value of Water Quality"; the conductivity determination is carried out in accordance with the standard GB / T 5750.4-2023 "Standard Test Methods for Drinking Water Part 4: Sensory Properties and Physical Indicators (9.1 Electrode Method)"; the dissolved oxygen determination is carried out in accordance with the standard HJ 506-2009 "Electrochemical Probe Method for Determination of Dissolved Oxygen in Water Quality"; the water temperature determination is carried out in accordance with the standard GB 13195-91 "Determination of Water Quality and Water Temperature"; the turbidity determination is carried out in accordance with the standard HJ 1075-2019 "Determination of Turbidity of Water Quality by Turbidimeter"; the algae density determination is carried out in accordance with the standard HJ1216-2021 "Determination of Phytoplankton in Water Quality by 0.1mL Counting Frame-Microscope Counting Method"; the chlorophyll a determination is carried out in accordance with the standard HJ 897-2017 "Water Quality - Determination of Chlorophyll a - Spectrophotometric Method" standard; the microcystin determination is carried out in accordance with GB / T 20466-2006 "Determination of Microcystins in Water" standard; the name of the sample tested and the measured value are recorded.

[0050] The fluorescence characteristics of the colored dissolved organic matter (CDOM) are similar to those of sodium humic acid (HA-Na) or fulvic acid (FA). In this case, sodium humic acid with a purity of 98% is used to replace the colored dissolved organic matter. Specifically, first, a certain mass of sodium humic acid is weighed, dissolved in ultrapure water to prepare a 1g / L stock solution, and stored in a brown bottle at 0-4°C. The dilution gradient is prepared on the spot, and the concentration of each gradient sodium humic acid solution is recorded and calculated, as shown in Table 1; then, a fluorescence spectrometer is used to scan and obtain the fluorescence peak intensity of each gradient sodium humic acid solution in Table 1, and a fluorescence peak intensity-concentration curve is drawn; finally, the fluorescence peak intensity of the sample solution to be tested is brought into the fluorescence peak intensity-concentration curve to obtain the concentration of sodium humic acid in the sample solution to be tested. The fluorescence scanning parameters were set as follows: excitation wavelength 351-660 nm, excitation wavelength interval 5 nm, excitation bandwidth 10 nm, emission wavelength 451-800 nm, emission bandwidth 10 nm, scanning interval 1.0 nm, and scanning speed 10000 nm / min.

[0051] Table 1

[0052] Serial number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 mg / L 0.2 0.4 0.6 0.8 1.0 1.5 2.0 2.5 3.0 3.5 4.0 5.0 10 15 20 25

[0053] (4) Fluorescence spectra of samples and test conditions

[0054] (401) Raman wavelength scanning is performed on the ultrapure water sample to obtain Raman wavelength scanning data of the ultrapure water. The Raman wavelength scanning parameters are set as follows: excitation wavelength 350 nm, excitation bandwidth 10 nm, emission wavelength scanning range 300-500 nm, emission bandwidth 10 nm, scanning interval 1.0 nm, and scanning speed 1000 nm / min.

[0055] (402) Performing three-dimensional fluorescence scanning on the ultrapure water sample, the algae liquid sample (single algae liquid sample and mixed algae liquid sample) prepared in step (2), the sample collected from natural water body and the sample containing colored dissolved organic matter, etc., to obtain fluorescence spectrum data of the corresponding samples. The three-dimensional fluorescence spectrum scanning parameters are set as follows: excitation wavelength scanning range 351-660nm, excitation wavelength interval 5nm, excitation bandwidth 10nm, emission wavelength scanning range 451-800nm, emission bandwidth 10nm, scanning interval 1.0nm, scanning speed 10000nm / min.

[0056] (5) Inversion of Microcystin LR

[0057] (501) Preprocessing of fluorescence data

[0058] ① Obtain fluorescence spectra of algae liquid and ultrapure water

[0059] Obtaining fluorescence spectrum data of algae liquid and ultrapure water according to step (4);

[0060] ② Sample minus blank

[0061] The blank subtraction refers to the difference between the fluorescence spectrum data matrix of the algae solution sample and the fluorescence spectrum data matrix of ultrapure water, denoted as M b ;

[0062] ③Raman unitization of fluorescence spectral data

[0063] The Raman unitization of the fluorescence spectrum data refers to the process of subtracting the blank fluorescence spectrum data M b Raman unitization is performed. Specifically, the fluorescence data of ultrapure water emission wavelength in the range of 371-421 nm is intercepted from the Raman wavelength scanning data obtained in step (401), and the Raman integral of the ultrapure water Raman peak is obtained after integration, which is recorded as The fluorescence spectrum data matrix M of the algae solution sample was deducted from the blank. b Divide by the Raman integral The new data obtained is the Raman unitized fluorescence spectrum data of the algae liquid sample, denoted as M la ;

[0064] The mathematical expression of Raman integration is:

[0065]

[0066] In the formula, represents the fluorescence intensity at the emission wavelength, λ em represents the emission wavelength, represents the starting emission wavelength, Indicates the final emission wavelength.

[0067] ④Deduct Rayleigh scattering and Raman scattering

[0068] The spectral data M of the fluorescence spectrum of the algae solution after Raman unitization la Subtract the interference of Rayleigh scattering and Raman scattering. Specifically, the influence of Rayleigh scattering is removed by setting the fluorescence intensity within the range of ±20-50nm of the excitation light wavelength to 0; Raman scattering is the inelastic scattering of water molecules, and its influence is eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; the Delaunay triangle interpolation method is used to interpolate the data to further remove or reduce the influence of Rayleigh scattering or Raman scattering, which is recorded as M d .

[0069] ⑤ Smoothing of spectral data

[0070] For data M d , use Gaussian kernel for smoothing to reduce noise interference, denoted as M g .

[0071] ⑥Selection and optimization of data matrix

[0072] The data matrix M g Specifically, when the algae liquid sample is scanned by fluorescence spectrometer, the excitation wavelength range of the instrument is set to 351-660nm, the scanning interval is 5nm, the emission wavelength range is 451-800nm, and the scanning interval is 1nm; therefore, the fluorescence spectrum data matrix obtained after scanning is 62×350, and M g After being converted into a one-dimensional array, the length of the array is 62×350=21700; denoted as M p ;

[0073] Preferably, selecting a suitable fluorescence spectrum range can not only enhance the recognition of algae species of various phyla and genera, but also further reduce interference to a certain extent, especially fluorescence interference from humus in natural water bodies. For this purpose, the data matrix M g , intercept the data of excitation wavelength 351-660nm and emission wavelength 661-760nm to form a matrix, convert it into a one-dimensional array, denoted as M y1 , whose length is 6200; the data of excitation wavelength 541-620nm and emission wavelength 622-661nm are intercepted to form a matrix, which is converted into a one-dimensional array and recorded as M y2, whose length is 640; the one-dimensional array M y1 and M y2 Merge into a new one-dimensional array, denoted as M p , the array length is 6840.

[0074] (502) Establishment of Ridge Regression Equation

[0075] ①Minimize the loss function

[0076] Establishing a quantitative relationship between fluorescence spectral data and microcystin LR

[0077] The (matrix) form of general linear regression analysis is as follows:

[0078]

[0079] In the above formula, x j is a one-dimensional array M p The elements in the equation are: y is the value of microcystin LR determined by HPLC, β is j is the slope or weight coefficient, and β0 is the intercept or constant term. The solution to the above equation is to use the least squares method to determine the β in the formula. j And β0, that is, the goal of the least squares method to solve the above regression problem is to minimize the following formula, that is, to minimize the loss function:

[0080]

[0081] in, is the predicted value of algae toxin or the value of algae toxin calculated by regression, y i is the regression dependent variable, and N is the number of samples.

[0082] ② Loss function of ridge regression regularization

[0083] The overfitting phenomenon caused by the model established by the least squares method, that is, the linear regression objective without regularization, can be corrected by the ridge regression model, that is, on the basis of the least squares method, L2 regularization is performed, and a penalty term equivalent to the square of the coefficient amplitude is added to constrain the model complexity, thereby reducing the risk of model overfitting and improving the generalization performance of the model.

[0084] Specifically, ridge regression adds a penalty term to the above minimization objective. Its mathematical expression is to add the sum of the squares of the weight vector Β (Beta) multiplied by a hyperparameter λ to the original loss function. The L2 penalty term of L2 regularization is inserted at the end of the loss function to form a new form, namely the ridge regression estimator, where λ is the regularization strength control parameter:

[0085]

[0086] By solving the above loss function with regularization term, we can get the coefficient estimate of ridge regression

[0087] The solution of the above equation can be automatically completed with the help of the Python learning library sklearn and the ridge regression (RidgeCV) module, or it can be completed with the help of corresponding modules in other programming languages.

[0088] ③ Establishment of regression model

[0089] For a series of samples (number of samples N), a one-dimensional array M of N samples is p Composition matrix X N , the matrix X N Each row represents the independent variable of a sample, and the matrix size is N×p.

[0090]

[0091] Similarly, for this series of samples (number of samples N), the microcystin LR measurement value y of each sample is composed into a matrix Y N , the matrix size is N×1.

[0092]

[0093] The matrix X N With the matrix Y N Form a new matrix M data , where Y N Located in the matrix M data The last column of the matrix M data After normalization, we get another new matrix, denoted as M sdata ;

[0094]

[0095] M sdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameter to build the model, and the model gives the predicted value of microcystin LR. The established model is denoted as M ridge .

[0096] Preferably, principal component analysis (PCA) is used to set the number of principal components and to analyze the matrix X N Perform dimensionality reduction processing, and the matrix after dimensionality reduction is recorded as XJN The number of principal components is set based on the premise of minimum information loss and the principle of representing most of the information of the original variables. The cumulative contribution rate of the principal components is usually not less than 80%.

[0097] The matrix X JN With the matrix Y N Form a new matrix M Jdata , where Y N Located in the matrix M Jdata The last column of the matrix M Jdata After normalization, we get another new matrix, denoted as M Jsdata ;

[0098] M Jsdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameter to build the model, and the model gives the predicted value of microcystin LR. The established model is denoted as M ridge .

[0099] ④Incorporate more water body measurement parameters as variables into the regression equation

[0100] Furthermore, in order to accurately invert the microcystin LR in algae water using fluorescence spectroscopy, it is also necessary to consider the factors that affect the measured fluorescence spectrum and the factors that affect the growth of algae, especially the factors that affect the growth of cyanobacteria.

[0101] The factors that usually affect fluorescence spectroscopy measurement and algae growth include: light, water temperature, salinity, total alkalinity, turbidity, pH value, colored dissolved organic matter (CDOM), dissolved oxygen (DO), etc.

[0102] Assume that while obtaining the fluorescence spectrum data of the sample, the pH, conductivity, dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, colored dissolved organic matter (CDOM) and other parameters of each sample are obtained, and the symbols a, b, c, d, e, f, g and h are used to replace these measured parameters. The measurement results of these parameters can be expressed in the following matrix:

[0103]

[0104] The matrix X N and the matrix W N Form a new matrix M XW , the matrix W N It is related to the parameters actually measured. The parameters that have not been measured are not included in the matrix W. N In the matrix M XWPerform principal component analysis to obtain the reduced dimension matrix M JXW , and then the matrix M JXW and the matrix Y N Form a new matrix M JXWY , where Y N Located in the matrix M JXWY The last column of the matrix M JXWY After normalization, we get the matrix M Jsdata .

[0105]

[0106] M Jsdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameter to build the model. The model then gives the predicted value of microcystin LR in each sample. The established model is denoted as M ridge .

[0107] (503) Prediction of Microcystin LR in Water Samples

[0108] Using the established regression model M ridge , the microcystin LR in the collected water samples to be tested can be predicted;

[0109] ① Obtain the fluorescence spectrum of the water sample and ultrapure water to be tested

[0110] According to step (4), obtaining fluorescence spectrum data of the water sample to be tested and ultrapure water;

[0111] ②Deduct blank space

[0112] Subtract the fluorescence spectrum data matrix of the water sample to be tested from the fluorescence spectrum data matrix of ultrapure water, denoted as M bx ;

[0113] ③Raman unitization of fluorescence spectral data

[0114] The Raman unitization of the fluorescence spectrum data refers to the process of subtracting the blank fluorescence spectrum data M bx Raman unitization is performed. Specifically, the fluorescence data of ultrapure water emission wavelength range 371-421nm is intercepted from the Raman wavelength scanning data obtained by (401), and the Raman integral of the ultrapure water Raman peak is obtained after integration, which is recorded as Using fluorescence spectrum data matrix M bx Divide by the Raman integral The new data obtained is the Raman unitized fluorescence spectrum data, denoted as M lax ;

[0115] The mathematical expression of Raman integration is:

[0116]

[0117] ④ Subtract Rayleigh scattering and Raman scattering of the water sample to be tested

[0118] The fluorescence spectrum M after Raman unitization lax Eliminate Rayleigh scattering and Raman scattering interference;

[0119] The effect of Rayleigh scattering is eliminated by setting the fluorescence intensity to 0 within the range of ±20-50nm of the excitation light wavelength.

[0120] Raman scattering is the inelastic scattering of water molecules, and its influence is eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water;

[0121] The data is interpolated using the Delaunay triangle interpolation method to further remove or reduce the influence of Rayleigh scattering or Raman scattering, denoted as M dx .

[0122] ⑤ Smoothing of spectral data

[0123] For data M dx , use Gaussian kernel for smoothing to reduce noise interference, denoted as M gx .

[0124] ⑥Selection and optimization of data matrix

[0125] The data matrix M gx Convert to a one-dimensional array;

[0126] Specifically, when the fluorescence spectrometer is used to scan the water sample to be tested, the excitation wavelength range of the instrument is set to 351-660nm, the scanning interval is 5nm, and the emission wavelength range is 451-800nm, and the scanning interval is 1nm; therefore, the fluorescence spectrum data matrix obtained after scanning is 62×350. gx After being converted into a one-dimensional array, the length of the array is 62×350=21700; denoted as M px .

[0127] Preferably, selecting a suitable fluorescence spectrum range can not only enhance the recognition of algae species of various phyla and genera, but also further reduce interference to a certain extent, especially fluorescence interference from humus in natural water bodies. For this purpose, the data matrix M gx , intercept the data of excitation wavelength 351-660nm and emission wavelength 661-760nm to form a matrix, convert it into a one-dimensional array, denoted as M yx1, whose length is 6200; the data of excitation wavelength 541-620nm and emission wavelength 622-661nm are intercepted to form a matrix, which is converted into a one-dimensional array and recorded as M yx2 , whose length is 640; the one-dimensional array M yx1 and M yx2 Merge into a new one-dimensional array, denoted as M px , the array length is 6840.

[0128] ⑦ Using the established model M ridge , predict the content of microcystin LR in the water sample to be tested

[0129] The one-dimensional array M of the water sample to be tested px Convert to matrix X p , and let X p X_Test, the predicted value of microcystin LR is Y_Pred, and the established model M ridge In the example, the model automatically gives the predicted value Y_Pred.

[0130] X p =|x1 x2…x p |

[0131] If the pH, conductivity, dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, colored dissolved organic matter (CDOM) and other parameters of the sample are obtained while obtaining the fluorescence spectrum data of the water sample to be tested, the symbols a, b, c, d, e, f, g and h are used to replace these measured parameters respectively, and the measurement results of these parameters can be expressed by the following matrix:

[0132] X p =|x1 x2…x p ab…h|

[0133] Let X p X_Test, the predicted value of microcystin LR is Y_Pred, and the established model M ridge In the example, the model automatically gives the predicted value Y_Pred.

[0134] Example 1

[0135] Microcystis aeruginosa was purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences and cultured according to step (1);

[0136] Take 20-30 mL of a qualified single algae species, Microcystis aeruginosa, and perform the original dilution. Specifically, dilute it to a fluorescence intensity of about 8000, which is the initial concentration value of this sample. Record the sample name OBS-A00-00-240411.

[0137] The initial sample was diluted to obtain 12 test samples;

[0138] According to step (3), the turbidity (expressed as e) and microcystin LR of the 12 test samples are measured;

[0139] According to steps (4) and (5), the obtained fluorescence spectrum data are processed. Specifically, the excitation wavelength range of the fluorescence spectrometer is set to 351-660nm, the scanning interval is 5nm, the emission wavelength range is set to 451-800nm, and the scanning interval is 1nm. The ultrapure water and the above 12 different concentrations of Microcystis aeruginosa water samples are scanned respectively. The obtained fluorescence spectrum data of Microcystis aeruginosa are deducted by blank, Raman unitization, Rayleigh scattering and Raman scattering are deducted in turn, and the spectrum data is smoothed to obtain a 62×350 fluorescence spectrum data matrix M g , M g After being converted into a one-dimensional array, the length of the array is 62×350=21700; denoted as M p , the M of 12 test samples p Form a new matrix X 12 , the matrix size is 12×21700, the matrix X 12 Each row of represents all the independent variables of a sample;

[0140] The turbidity data of each test sample is organized into the following matrix:

[0141]

[0142] The microcystin LR measurement value y of each sample in the table is composed of matrix Y 12 , the matrix size is 12×1;

[0143] The matrix X 12 and the matrix W 12 Form a new matrix M XW12 , for the matrix M XW12 Perform principal component analysis to obtain the reduced dimension matrix M JXW12 , and then the matrix M JXW12 and the matrix Y 12 Form a new matrix M JXWY12 , where Y 12 Located in the matrix M JXWY12 The last column of the matrix M JXWY12 After normalization, we get the matrix M Jsdata12 ;

[0144]

[0145] M Jsdata12The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameters to build the model. The established model is denoted as M ridge , and the model gives the predicted value of microcystin LR During the experiment, a model without adding turbidity data was also established, and the specific results are shown in the table.

[0146]

[0147] Example 2

[0148] A total of 11 species of algae from five phyla were purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences. The specific phyla and species names are as follows:

[0149]

[0150]

[0151] Cultivate according to step (1);

[0152] The above culture reaches the desired algae density of 10 7 cells / L or more, according to step (2), different algae species and algae genera are mixed, and the mixed algae genera are coded according to the above-mentioned phyla and species names, that is, the cyanobacteria of the same genus are coded A, and the same applies to the others;

[0153] Take the mixed algae liquid of the same algae genus as above, including cyanobacteria, green algae, diatoms, dinoflagellates, and cryptophytes (replaced by letters, i.e. A-cyanobacteria, B-green algae, C-diatoms, D-dinoflagellates, E-cryptophytes) and dilute them into 9 concentration gradients at 1:1.5 respectively;

[0154] The specific dilution gradient table is as follows:

[0155]

[0156] A total of 81 orthogonal mixed experimental groups were designed using SPSSAU;

[0157] The orthogonal experiment table is as follows:

[0158]

[0159]

[0160] Each experimental group is a certain volume of algae solution of the same algae genus, and the algae solution is cross-mixed according to the experimental design in the above orthogonal table, preferably with a volume ratio of A:B:C:D:E=1:1:1:1:1 to obtain the final mixed algae solution;

[0161] The 81 groups of mixed algae liquids were subjected to parameter measurement according to step (3), including turbidity (expressed by e), chlorophyll a (expressed by g), and microcystin LR. The concentration of chlorophyll a in the samples was measured to be distributed in the range of 0-600 μg / L, the turbidity was distributed in the range of 0-100 NTU, and the microcystin LR was distributed in the range of 0.5-15 μg / L.

[0162] The colored dissolved organic matter (CDOM) (represented by h) prepared in step (3) was added to the mixed algae solution prepared above, and the concentration of CDOM in each sample was measured to be distributed in the range of 0.6-10 mg / L;

[0163] Perform three-dimensional fluorescence spectrum scanning and data processing according to step (4);

[0164] According to the selection and optimization scheme of the data matrix in step (501), the preferred data matrix is ​​intercepted and converted into a one-dimensional array (i.e., the data matrix M g , intercept the data of excitation wavelength 351-660nm and emission wavelength 661-760nm to form a matrix, convert it into a one-dimensional array, denoted as M y1 , whose length is 6200; the data of excitation wavelength 541-620nm and emission wavelength 622-661nm are intercepted to form a matrix, which is converted into a one-dimensional array and recorded as M y2 , whose length is 640; the one-dimensional array M y1 and M y2 Merge into a new one-dimensional array, denoted as M p , the array length is 6840);

[0165] List X according to step (502) 81 matrix;

[0166] The turbidity, chlorophyll a and colored dissolved organic matter (CDOM) data for each test sample were organized into the following matrix:

[0167]

[0168] The microcystin LR measurement value y of each sample is composed into a matrix Y 81 , the matrix size is 81×1;

[0169] The matrix X 81 and the matrix W 81 Form a new matrix M XW81 , for the matrix M XW81 Perform principal component analysis to obtain the reduced dimension matrix M JXW81 , and then the matrix M JXW81 and the matrix Y 81 Form a new matrix MJXWY81 , where Y 81 Located in the matrix M JXWY81 The last column of the matrix M JXWY81 After normalization, we get the matrix M Jsdata81 ;

[0170]

[0171] M Jsdata81 The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameters to build the model. The established model is denoted as M ridge The model gives the predicted value of microcystin LR in each sample. The goodness of fit of the regression model R 2 >0.95.

[0172] Example 3

[0173] Five algae species purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences were cultured according to step 1;

[0174] The specific algae species and names are as follows:

[0175] Category Algae species name Latin name coding Cyanobacteria Microcystis aeruginosa Microcystis aeruginosa AA Chlorophyta Tetraselena obliquus Tetradesmus obliquus BB Diatoms Cyclotella meni Cyclotella meneghiniana CC Dinophyta Peridinium scutellariae Peridinium umbonatum var.inaequale DD Cryptophyta Cryptophytes Cryptomonas sp. EE

[0176] The fluorescence spectra of typical algae species and sodium humate are shown in Figure 1-6 .

[0177] The culture medium for the above algae species was BG11 medium for Cyanobacteria and Chlorophyta, CSI medium for Bacillariophyta, 119 medium for Dinophyta, and AF6 medium for Cryptophyta. The culture medium for each algae species was purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences.

[0178] The above culture reaches the desired algae density of 10 7 cells / L or more, mixing the five target algae species according to step (2), and finally obtaining 256 groups of mixed algae liquids;

[0179] Take the mixed algae liquid samples and measure the microcystin LR of each sample according to step (2);

[0180] The mixed algae liquid samples are used to measure the pH, conductivity, dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, colored dissolved organic matter (CDOM) and other parameters of each sample according to step (2). The parameters to be measured are represented by symbols a, b, c, d, e, f, g and h, respectively. The measurement results of these parameters are represented by the following matrix:

[0181]

[0182] Perform fluorescence spectrum scanning according to step (4); intercept the preferred data matrix according to the selection and optimization scheme of the data matrix in step (501), and convert it into a one-dimensional array;

[0183] List X in step (502) 256 Matrix, listing Y 256 matrix;

[0184] The matrix X 256 and the matrix W 256 Form a new matrix M XW256 , for the matrix M XW256 Perform principal component analysis to obtain the reduced dimension matrix M JXW256 , and then the matrix M JXW256 and the matrix Y 256 Form a new matrix M JXWY256 , where Y 256 Located in the matrix M JXWY256 The last column of the matrix M JXWY256 After normalization, we get the matrix M Jsdata256 ;

[0185]

[0186] M Jsdata256 The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross validation to automatically select the best regularization parameters to build the model. The established model is denoted as M ridge The model gives the predicted value of microcystin LR in each sample. The goodness of fit of the regression model R 2 >0.93.

[0187] Water samples were collected from natural water bodies, including Mahao Park (MHGY0), Dingjiahe (DJH00), and Dingjiahe Reservoir (DJHSK), and parameters such as fluorescence spectrum, pH, conductivity, dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, colored dissolved organic matter (CDOM) and microcystin LR were measured.

[0188] According to step (503), the water sample measurement data is substituted into the established model M ridge The content of microcystin LR in each water sample was obtained.

[0189] The data obtained by comparing the measured values ​​with the calculated values ​​are as follows:

[0190]

[0191] Note: ND means not detected;

[0192] LR calculation value before optimization: Ridge regression modeling was performed only based on the scanned fluorescence spectrum data information;

[0193] Optimized LR calculation value: Ridge regression modeling was performed based on the optimized fluorescence spectrum data information and influencing factors such as dissolved oxygen, water temperature, turbidity, algae density, chlorophyll a, and colored dissolved organic matter (CDOM).

Claims

1. A method for inverting the content of microcystin in water using algae fluorescence spectrum, characterized in that: The following steps are involved: (A1) Perform three-dimensional fluorescence scanning on N algae liquid samples of different concentrations to obtain the fluorescence spectrum data of the algae liquid samples, which are then converted into a one-dimensional array M of length p after preprocessing. p , the one-dimensional arrays corresponding to the N algae liquid samples are combined into a matrix X N , the matrix X N Each row of represents an independent variable of an algae liquid sample, and the matrix X N The size is N×p; (A2) Determine the microcystin content of the N algae liquid samples in step (A1) to form a matrix Y N , matrix Y N The size is N×1; (A3) Substitute the matrix X N With the matrix Y N Composition matrix M data , or the matrix X N After principal component analysis and dimensionality reduction, the matrix Y N Composition matrix M data , where Y N Located in the matrix M data The last column of the matrix M data Normalize and get the matrix M sdata , M sdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross-validation to automatically select the best regularization parameters to build a model. The established model is denoted as M ridge ; (A4) According to the processing method of step (A1), the fluorescence spectrum data of the sample to be tested is measured, and the fluorescence spectrum data is converted into a one-dimensional array M of length p after preprocessing. px , and then the one-dimensional array M px Convert to matrix X p , and let X p X_Test, the predicted value of microcystin content in the algae solution sample to be tested is Y_Pred, and it is substituted into the established model M ridge In the example, the model automatically gives the predicted value Y_Pred.

2. A method for inverting the content of microcystin in water using algae fluorescence spectrum, characterized in that: The following steps are involved: (B1) Perform three-dimensional fluorescence scanning on N algae liquid samples of different concentrations to obtain the fluorescence spectrum data of the algae liquid samples, which are converted into a one-dimensional array M of length p after preprocessing. p , the one-dimensional arrays corresponding to the N algae liquid samples are combined into a matrix X N , the matrix X N Each row of represents an independent variable of an algae liquid sample, and the matrix X N The size is N×p. The influencing factors of N algae liquid samples are collected at the same time and expressed in matrix W N Indicates that the matrix W N The size is N×m, where m is the number of impact factors; (B2) Determine the microcystin content of N algae samples with different concentrations to form a matrix Y N , the matrix size is N×1; (B3) Substitute the matrix X N and the matrix W N Composition matrix M XW , the matrix M XW The size is N×(p+m), for the matrix M XW Perform principal component analysis to obtain the reduced dimension matrix M JXW , and then the matrix M JXW and the matrix Y N Composition matrix M JXWY , where Y N Located in the matrix M JXWY The last column of the matrix M JXWY After normalization, we get the matrix M Jsdata , M Jsdata The last column in is denoted as Y_Train, and the rest are denoted as X_Train. Substitute them into RidgeCV and use cross-validation to automatically select the best regularization parameters to build a model. The established model is denoted as M ridge ; (B4) According to the processing method of step (B1), the fluorescence spectrum data of the sample to be tested is measured, and the fluorescence spectrum data is converted into a one-dimensional array M of length p after preprocessing. px , and determine the influencing factors of the samples to be tested, and convert the one-dimensional array M px The transformed matrix and the matrix of influencing factors are combined into the matrix X p , and let X p X_Test, the predicted value of microcystin LR is Y_Pred, and the established model M ridge In the example, the model automatically gives the predicted value Y_Pred.

3. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 1 or 2, characterized in that: The algae solution samples in steps (A1) and (B1) and the samples to be tested in steps (A4) and (B4) include single algae solution samples and mixed algae solution samples, and steps (A1) and (A4), steps (B1) and (B4) should correspond to each other. The preparation of single algae solution refers to taking a certain volume of algae species and culturing them to reach the required algae density of 10 7 cells / L or more, and a series of single algae solutions of different concentrations are obtained by diluting the algae solution of a certain algae genus of a certain algae phylum according to a certain dilution gradient; the preparation of mixed algae solution samples refers to taking a certain volume of algae of different algae genera of the same algae phylum that have reached the required algae density of 10 7 cells / L or more, by mixing or diluting algae in a certain proportion to obtain a series of mixed algae solutions of various concentrations; it may also refer to a series of mixed algae solutions of various concentrations obtained by mixing or diluting a certain volume of algae from different algae genera and phyla in a certain proportion to obtain a series of mixed algae solutions of various concentrations.

4. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 1 or 2, characterized in that: The parameters for the three-dimensional fluorescence scanning measurement in steps (A1) and (B1) are: excitation wavelength scanning range 351-660nm, excitation wavelength interval 5nm, excitation bandwidth 10nm, emission wavelength scanning range 451-800nm, emission bandwidth 10nm, scanning interval 1.0nm, and scanning speed 10000nm / min.

5. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 4, characterized in that: The fluorescence spectrum data matrix obtained after three-dimensional fluorescence scanning is converted into a one-dimensional array M p .

6. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 4, characterized in that: For the fluorescence spectrum data matrix obtained after three-dimensional fluorescence scanning, the data of excitation wavelength 351-660nm and emission wavelength 661-760nm are intercepted to form a matrix and converted into a one-dimensional array M y1 ; In addition, the data of excitation wavelength 541-620nm and emission wavelength 622-661nm are intercepted to form a matrix and converted into a one-dimensional array M y2 ; The one-dimensional array M y1 and M y2 Merge into a new one-dimensional array M p .

7. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 1 or 2, characterized in that: In steps (A2) and (B2), the content of microcystin is determined according to GB / T 20466-2006 "Determination of Microcystin in Water".

8. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 1 or 2, characterized in that: The influencing factors in steps (B1) and (B4) include one or more of light, water temperature, salinity, total alkalinity, turbidity, pH value, colored soluble organic matter, and dissolved oxygen.

9. The method for inverting the content of microcystin in water using algae fluorescence spectrum according to claim 1 or 2, characterized in that: The preprocessing includes blank subtraction, Raman unitization, Rayleigh scattering and Raman scattering subtraction, data smoothing and data matrix selection and optimization, and specifically includes the following steps: (101) subtracting the difference between the fluorescence spectrum data matrix of the algae liquid sample or the sample to be tested and the fluorescence spectrum data matrix of the ultrapure water; (102) Based on the obtained Raman wavelength scanning data, the fluorescence data of ultrapure water emission wavelength range of 371-421 nm was intercepted and integrated to obtain the Raman integral of the ultrapure water Raman peak, which is denoted as Divide the fluorescence spectrum data matrix of the algae solution sample or the sample to be tested by the Raman integral after subtracting the blank The new data obtained are Raman unitized fluorescence spectrum data of the algae liquid sample or the sample to be tested; (103) The influence of Rayleigh scattering is removed by setting the fluorescence intensity within the range of ±20-50 nm of the excitation light wavelength to zero; Raman scattering is the inelastic scattering of water molecules, and its influence is eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; the Delaunay triangle interpolation method is used to interpolate the data to further remove or reduce the influence of Rayleigh scattering or Raman scattering; (104) Smoothing the data obtained in step (103) using a Gaussian kernel to reduce noise interference; (105) Convert the data matrix obtained in step (104) into a one-dimensional array M p , or M px .

Images