A method for retrieving microcystin content in water by using algal fluorescence spectrum

Abstract

This invention discloses a method for retrieving microcystin content in water using algal fluorescence spectroscopy. First, fluorescence spectral data of N algal solution samples with different concentrations are measured, and after preprocessing, the data are converted into a one-dimensional array M of length p. p Furthermore, all the corresponding one-dimensional arrays are combined into a matrix X. N The microcystin content of N algal solution samples was determined, forming a matrix Y. N Based on matrix X N With matrix Y N Establish model M ridge The fluorescence spectrum data of the sample to be tested were measured, and after preprocessing, they were converted into a one-dimensional array M of length p. px This is further transformed into matrix X. p Substitute into the established model M ridge In the model, the predicted values ​​of microcystin are automatically provided. Model M ridge During the modeling process, factors affecting fluorescence spectroscopy and algal growth can be incorporated to further improve model accuracy. This method is more accurate, simple, quick, and convenient, facilitating on-site monitoring and possessing significant practical value.

Description

Technical Field

[0001] This invention belongs to the field of water environment pollutant detection technology, specifically relating to a method for inverting the content of microcystin-LR in water using algal fluorescence spectroscopy. Technical Background

[0002] With the development of technology, more and more water bodies are severely polluted, leading to algal blooms and red tides in many rivers, lakes, and oceans. The most representative algal bloom is cyanobacterial bloom, where the dying algae release large amounts of microcystins. Microcystins are among the most common and toxic toxins found in freshwater. They are cyclic peptide hepatotoxins that pose significant risks to animal and human health. The production of microcystins not only directly affects drinking water safety but also causes ecosystem imbalances. Furthermore, they accumulate in the human body through plant and animal growth, leading to health problems, promoting tumor development, poisoning, and even death. Therefore, real-time monitoring of microcystins in water bodies has always been an important research topic. As an important indicator for drinking water, the limit for microcystins is no more than 1.0 μg / L.

[0003] Currently, the main methods for monitoring algal toxins include chemical methods, biological methods, and immunoassay methods. Chemical methods utilize instruments such as high-performance liquid chromatography (HPLC). These methods are expensive, involve complex and time-consuming pretreatment steps, and are difficult to implement in real-time monitoring, especially for low concentrations of MC-LR, where significant errors often occur. Biological methods primarily measure acute toxicity in animals, usually through intravenous injection or gavage in mice, generally using the median lethal dose (LD50). This method is simple and crude but cannot provide precise quantification. Immunoassay methods use enzyme-linked immunosorbent assay (ELISA) readers to detect the binding of specific antigens and antibodies, utilizing specific markers for identification. This method is highly sensitive, selective, stable, and efficient, but it is highly specialized, time-consuming, and does not provide immediate results.

[0004] Patent CN 113029994 A discloses a method for inverting microcystin concentration based on multi-source characteristic spectra of extracellular organic matter. It establishes the relationship between chlorophyll a from the multi-source characteristic spectra of extracellular organic matter in *Microcystis aeruginosa* and microcystins. However, in practice, this method has been found to fail to consider the complexity of actual natural water bodies. For example, humus and turbidity in natural water bodies can interfere with the fluorescence measurement of algae, causing a significant deviation between predicted and actual values, making accurate quantification difficult. Summary of the Invention

[0005] The purpose of this invention is to provide a method for inverting the content of microcystin in water using algal fluorescence spectroscopy, which solves the problem that existing methods do not take into account the complexity of actual natural water bodies, resulting in serious deviations between predicted and actual values ​​and low accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for retrieving microcystin content in water using algal fluorescence spectroscopy includes the following steps:

[0008] (A1) Three-dimensional fluorescence scanning was performed on N algal solution samples with different concentrations to obtain the fluorescence spectrum data of the algal solution samples. After preprocessing, the data were converted into a one-dimensional array M of length p. p The one-dimensional arrays corresponding to the N algal liquid samples are combined to form a matrix X. N Matrix X N Each row represents an independent variable for an algal solution sample, matrix X N The size is N×p;

[0009] (A2) The microcystin content of the N algal liquid samples described in step (A1) is determined to form a matrix Y. N Matrix Y N The size is N×1;

[0010] (A3) Matrix X N With matrix Y N Composition matrix M data Or matrix X N Principal component analysis dimensionality reduction and matrix Y N Composition matrix M data , where Y N Located in matrix M data The last column, for matrix M data Normalization is performed to obtain matrix M sdata M sdata The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge ;

[0011] (A4) Following the processing method in step (A1), the fluorescence spectrum data of the sample to be tested is measured, and after preprocessing, it is converted into a one-dimensional array M of length p. px Then, the one-dimensional array M px Convert to matrix X p And let X p Let X_Test be the predicted value of microcystin content in the algal solution sample, and let Y_Pred be the predicted value. Substitute this value into the model M established in step (A3). ridge In this process, the model automatically provides the predicted value Y_Pred.

[0012] Accurately retrieving microcystin levels in algal-rich water using fluorescence spectroscopy requires considering factors affecting fluorescence spectroscopy and algal growth (collectively referred to as influencing factors), such as light intensity, water temperature, salinity, total alkalinity, turbidity, pH, colored dissolved organic matter (CDOM), and dissolved oxygen (DO). Therefore, an optimized method for retrieving microcystin (LR) levels in water using algal fluorescence spectroscopy includes the following steps:

[0013] (B1) Three-dimensional fluorescence scanning was performed on N algal solution samples with different concentrations to obtain the fluorescence spectrum data of the algal solution samples. After preprocessing, the data were converted into a one-dimensional array M of length p. p The one-dimensional arrays corresponding to N algal liquid samples are combined to form a matrix X. N Matrix X N Each row represents an independent variable for an algal solution sample, matrix X N The influence factors of N×p algal liquid samples are collected simultaneously and represented by matrix W. N Indicate that matrix W N The size is N×m, where m is the number of influencing factors;

[0014] (B2) The microcystin content of N algal solution samples with different concentrations was determined, and matrix Y was constructed. N The matrix size is N×1;

[0015] (B3) Matrix X N Sum matrix W N Composition matrix M XW Matrix M XW For a matrix M of size N×(p+m), XW Principal component analysis was performed to obtain the dimension-reduced matrix M. JXW Then matrix M JXW And matrix Y N Composition matrix M JXWY , where Y N Located in matrix M JXWY The last column, for matrix M JXWY After normalization, matrix M is obtained. Jsdata M Jsdata The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge ;

[0016] (B4) Following the processing method in step (B1), the fluorescence spectrum data of the sample to be tested is measured, and after preprocessing, it is converted into a one-dimensional array M of length p. px Simultaneously, the influence factors of the sample to be tested are determined, and a one-dimensional array M is used.px The transformed matrix and the matrix composed of the influence factors are combined to form matrix X. p And let X p Let X_Test be the microcystin LR prediction value and Y_Pred be the value. Substitute this value into the model M already established in step (B3). ridge In this process, the model automatically provides the predicted value Y_Pred.

[0017] The algal solution samples in steps (A1) and (B1) and the test samples in steps (A4) and (B4) all include single algal solution samples and mixed algal solution samples, and steps (A1) and (A4), and steps (B1) and (B4) should correspond consistently. Specifically, the preparation of single algal solution refers to taking a certain volume of algal seed and culturing it to reach the required algal density of 10. 7 The preparation of mixed algal solutions involves taking a certain volume of algal solutions from a specific phylum and genus of algae with a density of 10⁻⁶ cells / L or higher, and diluting them according to a specific dilution gradient to obtain a series of single algal solutions of different concentrations; the preparation of mixed algal solution samples refers to taking a certain volume of different genera of the same algal phylum that have reached the desired algal density of 10⁻⁶ cells / L. 7 Algae with a cell / L or higher concentration are mixed or diluted in a certain proportion to obtain a series of mixed algal solutions of various concentrations; or they can refer to a series of mixed algal solutions of various concentrations obtained by mixing or diluting algae of different phyla and genera in a certain proportion to obtain a certain volume of algae.

[0018] The parameters for the three-dimensional fluorescence scanning measurement in steps (A1) and (B1) are 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, and scanning speed 10000nm / min.

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

[0020] As another preferred method, for the fluorescence spectral data matrix obtained after three-dimensional fluorescence scanning, the data with excitation wavelengths of 351-660 nm and emission wavelengths of 661-760 nm are truncated to form a matrix, which is then converted into a one-dimensional array M. y1 Furthermore, data from excitation wavelengths of 541-620nm and emission wavelengths of 622-661nm were extracted to form a matrix, which was then converted into a one-dimensional array M. y2 ; Transform the one-dimensional array M y1 and M y2 Merge into a one-dimensional array M p .

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

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

[0023] The preprocessing in steps (A1) and (B1) sequentially includes blank removal, Raman normalization, Rayleigh and Raman scattering subtraction, data smoothing, and data matrix selection and optimization. Specifically, it includes the following steps:

[0024] (101) Subtract the difference between the fluorescence spectral data matrix of the algal solution sample or the sample to be tested and the fluorescence spectral data matrix of the ultrapure water.

[0025] (102) From the obtained Raman wavelength scanning data, the fluorescence data in the emission wavelength range of ultrapure water (371-421 nm) is extracted, integrated, and the Raman integral of the ultrapure water Raman peak is obtained, denoted as . Divide the fluorescence spectral data matrix of the algal solution sample (excluding the blank) or the sample to be tested by the Raman integral. The new data obtained are Raman unitized fluorescence spectra of algal liquid samples or samples to be tested;

[0026] (103) The influence of Rayleigh scattering was removed by setting the fluorescence intensity to 0 within the range of ±20-50nm of the excitation wavelength; Raman scattering is an inelastic scattering of water molecules, and its influence was eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; the data were interpolated using the Delaunay triangle interpolation method to further remove or reduce the influence of Rayleigh scattering or Raman scattering.

[0027] (104) The data obtained in step (103) is smoothed 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] This invention utilizes algal fluorescence spectroscopy to invert the content of microcystin in water. It obtains three-dimensional fluorescence data by scanning water bodies containing algae and uses water quality parameters affecting fluorescence measurement, such as chlorophyll a, algal density, turbidity, and colored dissolved organic matter, as influencing factors. A ridge regression algorithm is used to directly construct a multiple linear relationship between the three-dimensional fluorescence spectral data and microcystin, and a model is built to calculate the content of microcystin-LR in the water body. This method is more accurate, simpler, faster, and more convenient, facilitating on-site monitoring and possessing significant practical value. Attached Figure Description

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

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

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

[0033] Figure 4 This is a fluorescence spectrum of dinoflagellates.

[0034] Figure 5 This is a fluorescence spectrum of Cryptophyte.

[0035] Figure 6 This is the fluorescence spectrum of sodium humate. Detailed Implementation

[0036] To make the objectives and technical solutions of this invention clearer, the technical solutions of this invention will be further described below in conjunction with embodiments, but the description of this invention is not limited to the following.

[0037] Example

[0038] A method for retrieving microcystin (LR) content in water using algal fluorescence spectroscopy

[0039] (1) Algal culture

[0040] (101) Five dominant algae from five common phyla in my country’s water bodies were selected as the subjects of the experiment. The dominant algae include cyanobacteria, green algae, diatoms, dinoflagellates and cryptophytes, and all were purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences.

[0041] (102) The purchased algal strains were processed according to their instructions. The algal solution in the test tubes was shaken well and transferred to sterile glass Erlenmeyer flasks in a clean bench. The flasks were then placed in a smart constant temperature and humidity incubator under low light for 2-3 days. The incubator water temperature was set at 25℃±1℃, the light intensity at 1000-4000 lux, and the light duration at 12h day / 12h night. When the algal strains showed good growth and a significant increase in biomass, they were transferred again to the corresponding culture medium at a ratio of 1:5. The Erlenmeyer flasks were shaken 2-3 times a day, and the position of the flasks was adjusted as needed to ensure uniform light. Microscopic examination was performed 1-2 times per week. BG11 medium was used for cyanobacteria and green algae, CSI medium for diatoms, 119 medium for dinoflagellates, and AF6 medium for cryptophytes. All culture media were purchased from the Freshwater Algae Culture Collection of the Chinese Academy of Sciences.

[0042] (103) When the above-mentioned algae growth reaches an algal density of 10 7When the algal density is above cells / L or the fluorescence intensity of the algal solution measured by the fluorescence spectrometer reaches about 10,000, proceed to step (2). The algal density is counted according to the standard of HJ 1216-2021 "Determination of Phytoplankton in Water Quality - Microscopic Counting Method with 0.1mL Counting Frame".

[0043] (2) Preparation of algal solution

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

[0045] The aforementioned preparation of a single algal solution refers to culturing a certain volume of algal seed until the desired algal density of 10⁻⁶ is reached. 7 A series of single algal solutions of different concentrations are obtained by diluting a single algal species of a certain algal phylum with a cell / L or higher according to a certain dilution gradient.

[0046] The preparation of the mixed algal solution refers to taking a certain volume of different genera of the same algal phylum that have reached the required algal density of 10. 7 Algae with a cell / L or higher concentration are mixed or diluted in a certain proportion to obtain a series of mixed algal solutions of various concentrations; or they can refer to a series of mixed algal solutions of various concentrations obtained by mixing or diluting algae of different phyla and genera in a certain proportion.

[0047] (202) Take a certain volume of the above-mentioned material to reach the required algal density of 10. 7 Various algal solutions with a cell / L or higher were diluted to approximately 60%-80% of the fluorescence intensity of the initial concentration.

[0048] (3) Measurement of various parameters

[0049] Take a certain volume of single or mixed algal solutions of different concentrations prepared in step (2) as samples, and complete the determination of parameters such as pH, conductivity, dissolved oxygen, water temperature, turbidity, algal density, chlorophyll a, colored dissolved organic matter (CDOM), and microcystin. The pH measurement was performed according to HJ 1147-2020 "Determination of pH Value in Water - Electrode Method"; the conductivity measurement was performed according to GB / T 5750.4-2023 "Standard Examination Methods for Drinking Water - Part 4: Sensory Characteristics and Physical Indicators (9.1 Electrode Method)"; the dissolved oxygen measurement was performed according to HJ 506-2009 "Determination of Dissolved Oxygen in Water - Electrochemical Probe Method"; the water temperature measurement was performed according to GB 13195-91 "Determination of Water Temperature in Water"; the turbidity measurement was performed according to HJ 1075-2019 "Determination of Turbidity in Water - Turbidity Meter Method"; the algae density measurement was performed according to HJ1216-2021 "Determination of Phytoplankton in Water - 0.1mL Counting Frame - Microscopic Counting Method"; and the chlorophyll a measurement was performed according to HJ... The determination of microcystin a in water was performed according to the standard GB / T 20466-2006 "Determination of Microcystin in Water"; the sample name and measured value were recorded.

[0050] The fluorescence characteristics of the colored dissolved organic matter (CDOM) are similar to those of sodium humate (HA-Na) or fulvic acid (FA). In this case, sodium humate with a purity of 98% is used to replace the colored dissolved organic matter. Specifically: First, a certain mass of sodium humate is weighed and dissolved in ultrapure water to prepare a 1 g / L stock solution, which is then refrigerated in a brown bottle at 0-4℃. The dilution gradient is prepared fresh for each use, and the concentration of sodium humate solution at each gradient is recorded and calculated, as shown in Table 1. Then, the fluorescence peak intensity of sodium humate solution at each gradient is obtained by scanning with a fluorescence spectrometer, and a fluorescence peak intensity-concentration curve is plotted. Finally, the fluorescence peak intensity of the sample solution to be tested is substituted into the fluorescence peak intensity-concentration curve to obtain the concentration of sodium humate in the sample solution to be tested. The fluorescence scanning parameters were set as follows: excitation wavelength 351-660nm, excitation wavelength interval 5nm, excitation bandwidth 10nm, emission wavelength 451-800nm, emission bandwidth 10nm, scan interval 1.0nm, and scan speed 10000nm / 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 spectrum of the sample and test conditions

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

[0055] (402) Three-dimensional fluorescence scanning was performed on ultrapure water samples, algal solution samples prepared in step (2) (single algal solution samples and mixed algal solution samples), samples collected from natural water bodies, and samples containing colored dissolved organic matter, to obtain the corresponding fluorescence spectral data. The three-dimensional fluorescence spectral scanning parameters were 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, and scanning speed 10000nm / min.

[0056] (5) Inversion of microcystin LR

[0057] (501) Preprocessing of fluorescence data

[0058] ① Obtain fluorescence spectra of algal solutions and ultrapure water

[0059] Obtain fluorescence spectral data of algal solution and ultrapure water according to step (4);

[0060] ② Sample subtraction blank

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

[0062] ③ Raman normalization of fluorescence spectral data

[0063] Raman normalization of the fluorescence spectral data refers to the process of normalizing the fluorescence spectral data M after subtracting the blank. b Raman normalization is performed. Specifically, from the Raman wavelength scan data obtained in step (401), fluorescence data in the emission wavelength range of 371-421 nm of ultrapure water is extracted, integrated, and the Raman integral of the ultrapure water Raman peak is obtained, denoted as . Fluorescence spectral data matrix M of algal solution samples after subtracting blanks b Divide by Raman integral The new data obtained are Raman normalized fluorescence spectra of algal solutions, denoted as M. la ;

[0064] The mathematical expression for the Raman integral is:

[0065]

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

[0067] ④ Subtract Rayleigh scattering and Raman scattering

[0068] M, the Raman-normalized spectral data of the algal solution fluorescence spectrum la To eliminate interference from Rayleigh and Raman scattering, the following steps were taken: Rayleigh scattering was removed by setting the fluorescence intensity to 0 within the ±20-50 nm range of the excitation wavelength; Raman scattering, being an inelastic scattering of water molecules, was eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; and the Delaunay triangle interpolation method was used to further remove or reduce the effects of Rayleigh or Raman scattering, denoted as M. d .

[0069] ⑤ Smoothing of spectral data

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

[0071] ⑥ Selection and optimization of data matrices

[0072] Data matrix M g Convert to a one-dimensional array. Specifically, when scanning algal samples using a fluorescence spectrometer, the instrument is set with an excitation wavelength range of 351-660 nm and a scan interval of 5 nm, and an emission wavelength range of 451-800 nm and a scan interval of 1 nm; therefore, the resulting fluorescence spectral data matrix is ​​62×350. g After being converted into a one-dimensional array, the array length is 62 × 350 = 21700; denoted as M. p ;

[0073] Preferably, selecting an appropriate fluorescence spectral range can not only enhance the identification of algal species of various phyla and genera, but also further reduce interference to a certain extent, especially fluorescence interference from humic substances in natural water bodies. Therefore, the data matrix M... g Data from excitation wavelengths of 351-660nm and emission wavelengths of 661-760nm are extracted and formed into a matrix, which is then converted into a one-dimensional array, denoted as M. y1 Its length is 6200; additionally, data from excitation wavelengths of 541-620nm and emission wavelengths of 622-661nm are extracted to form a matrix, which is converted into a one-dimensional array denoted as M. y2Its length is 640; the one-dimensional array M y1 and M y2 Merge them into a new one-dimensional array, denoted as M. p The array length is 6840.

[0074] (502) Establishment of Ridge Regression Equation

[0075] ① Minimize the loss function

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

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

[0078]

[0079] x in the above formula j For a one-dimensional array M p The elements in the figure, y is the LR value of microcystin obtained by high performance liquid chromatography, and β j β is the slope or weighting coefficient, and β0 is the intercept or constant term. Solving the above equation involves using the least squares method to determine β in the formula. j And β0, the goal of solving the above regression problem using the least squares method is to minimize the following expression, that is, to minimize the loss function:

[0080]

[0081] in, y represents the predicted or regression-calculated value of algal toxins. i For regression, the dependent variable is N, which refers to the sample size.

[0082] ②Loss function of ridge regression regularization

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

[0084] Specifically, ridge regression adds a penalty term to the aforementioned minimization objective. Mathematically, it is expressed as the sum of squares of the weight vector B (beta) multiplied by a hyperparameter λ, added to the original loss function. The L2 regularization penalty term is inserted at the end of the loss function, forming a new form, the ridge regression estimator, where λ is the regularization strength control parameter.

[0085]

[0086] By solving the loss function with regularization, the coefficient estimates of the ridge regression are obtained.

[0087] The above equations can be solved automatically using the Ridge Regression (RidgeCV) module imported from the Python learning library sklearn, or with the help of corresponding modules in other programming languages.

[0088] ③ Establishment of the regression model

[0089] For a series of samples (number of samples N), the one-dimensional array M of the N samples is... p Composition matrix X N 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 values ​​y of each sample are used to form a matrix Y. N The matrix size is N×1.

[0092]

[0093] matrix X N With matrix Y N Form a new matrix M data , where Y N Located in matrix M data The last column, for matrix M data After normalization, we obtain a new matrix, denoted as M. sdata ;

[0094]

[0095] M sdata The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model, which then provides the predicted values ​​for 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 for matrix X. N Dimensionality reduction is performed, and the resulting matrix is ​​denoted as X.JN The number of principal components should be set with the premise of minimal information loss and the principle of representing most of the information of the original variables. Usually, the cumulative contribution rate of the principal components should not be less than 80%.

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

[0098] M Jsdata The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model, which then provides the predicted values ​​for microcystin LR. The established model is denoted as M. ridge .

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

[0100] Furthermore, to accurately retrieve the LR of microcystin in algal water using fluorescence spectroscopy, it is necessary to consider factors that affect the measurement of fluorescence spectra and factors that affect algal growth, especially factors that affect the growth of cyanobacteria.

[0101] Factors that typically affect fluorescence spectroscopy measurements and algal growth include: light, water temperature, salinity, total alkalinity, turbidity, pH value, colored dissolved organic matter (CDOM), and dissolved oxygen (DO).

[0102] Suppose that while acquiring the fluorescence spectrum data of the samples, parameters such as pH, conductivity, dissolved oxygen, water temperature, turbidity, algal density, chlorophyll a, and colored dissolved organic matter (CDOM) for each sample are also acquired. These parameters are represented by the symbols a, b, c, d, e, f, g, and h, respectively. The measurement results for these parameters can be represented by the following matrix:

[0103]

[0104] matrix X N Sum matrix W N Form a new matrix M XW Matrix W N Parameters that are not measured are not included in matrix W. N In the matrix M XWPrincipal component analysis was performed to obtain the dimension-reduced matrix M. JXW Then matrix M JXW And matrix Y N Form a new matrix M JXWY , where Y N Located in matrix M JXWY The last column, for matrix M JXWY After normalization, matrix M is obtained. Jsdata .

[0105]

[0106] M Jsdata The last column is denoted as Y_Train, and the rest as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The model then provides a predicted value for microcystin LR in each sample. The established model is denoted as M. ridge .

[0107] (503) Prediction of microcystin LR in water samples to be tested

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

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

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

[0111] ②Deduct blanks

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

[0113] ③ Raman normalization of fluorescence spectral data

[0114] Raman normalization of the fluorescence spectral data refers to the process of normalizing the fluorescence spectral data M after subtracting the blank. bx Raman normalization is performed. Specifically, from the Raman wavelength scan data obtained by (401), fluorescence data in the emission wavelength range of ultrapure water (371-421 nm) is extracted, integrated, and the Raman integral of the ultrapure water Raman peak is obtained, denoted as . Using fluorescence spectroscopy data matrix M bx Divide by Raman integral The new data obtained are Raman-normalized fluorescence spectral data, denoted as M. lax ;

[0115] The mathematical expression for the Raman integral is:

[0116]

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

[0118] The fluorescence spectrum M after Raman normalization lax Subtract Rayleigh scattering and Raman scattering interference;

[0119] The influence of Rayleigh scattering was removed by setting the fluorescence intensity to 0 within the range of ±20-50 nm of the excitation wavelength;

[0120] Raman scattering is an inelastic scattering of water molecules, and its influence can be 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 effects of Rayleigh or Raman scattering, denoted as M. dx .

[0122] ⑤ Smoothing of spectral data

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

[0124] ⑥ Selection and optimization of data matrices

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

[0126] Specifically, when scanning the water sample using a fluorescence spectrometer, the instrument was set with an excitation wavelength range of 351-660 nm and a scanning interval of 5 nm, and an emission wavelength range of 451-800 nm and a scanning interval of 1 nm; therefore, the resulting fluorescence spectral data matrix was 62×350. gx After being converted into a one-dimensional array, the array length is 62 × 350 = 21700; denoted as M. px .

[0127] Preferably, selecting an appropriate fluorescence spectral range can not only enhance the identification of algal species of various phyla and genera, but also further reduce interference to a certain extent, especially fluorescence interference from humic substances in natural water bodies. Therefore, the data matrix M... gx Data from excitation wavelengths of 351-660nm and emission wavelengths of 661-760nm are extracted and formed into a matrix, which is then converted into a one-dimensional array, denoted as M. yx1Its length is 6200; additionally, data from excitation wavelengths of 541-620nm and emission wavelengths of 622-661nm are extracted to form a matrix, which is converted into a one-dimensional array denoted as M. yx2 Its length is 640; the one-dimensional array M yx1 and M yx2 Merge them 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 Let X_Test be the LR prediction value of microcystin, and Y_Pred be the value of the microcystin LR prediction. Substitute this value into the established model M. ridge In this process, the model automatically provides the predicted value Y_Pred.

[0130] X p =|x1 x2…x p |

[0131] If, while acquiring the fluorescence spectrum data of the water sample to be tested, parameters such as pH, conductivity, dissolved oxygen, water temperature, turbidity, algal density, chlorophyll a, and colored dissolved organic matter (CDOM) are also acquired, and these parameters are represented by the symbols a, b, c, d, e, f, g, and h, respectively, the measurement results of these parameters can be represented by the following matrix:

[0132] X p =|x1 x2…x p ab…h|

[0133] Let X p Let X_Test be the LR prediction value of microcystin, and Y_Pred be the value of the microcystin LR prediction. Substitute this value into the established model M. ridge In this process, the model automatically provides 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 single, successfully cultured cyanobacterial species, Microcystis aeruginosa, and perform an initial dilution. Specifically, dilute to a fluorescence intensity of approximately 8000. This is the initial concentration value of the sample. Record the sample name as OBS-A00-00-240411.

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

[0138] The turbidity (represented by e) and microcystin LR of the above 12 test samples were determined according to step (3);

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

[0140] The turbidity data for each test sample were compiled into the following matrix:

[0141]

[0142] The microcystin LR measurement values ​​y for each sample in the table are used to construct a matrix Y. 12 The matrix size is 12×1;

[0143] matrix X 12 Sum matrix W 12 Form a new matrix M XW12 For matrix M XW12 Principal component analysis was performed to obtain the dimension-reduced matrix M. JXW12 Then matrix M JXW12 And matrix Y 12 Form a new matrix M JXWY12 , where Y 12 Located in matrix M JXWY12 The last column, for matrix M JXWY12 After normalization, matrix M is obtained. Jsdata12 ;

[0144]

[0145] M Jsdata12The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge The model provides predicted values ​​for microcystin LR. A model without turbidity data was also established during the experiment, and the specific results are shown in the table.

[0146]

[0147] Example 2

[0148] Eleven species and genera from five phyla were purchased from the Freshwater Algae Seed Bank of the Chinese Academy of Sciences. The specific phyla and genera are as follows:

[0149]

[0150]

[0151] Cultivate according to step (1);

[0152] The above culture was carried out to achieve the desired algal density of 10. 7 Eleven algal species with cells / L or higher were mixed with each other within the same genus according to step (2). The different genus of algae after mixing were coded according to the above phylum and species name. That is, the cyanobacteria of the same genus were coded as A, and the others were coded in the same way.

[0153] Take the above-mentioned mixed algal solutions of cyanobacteria, green algae, diatoms, dinoflagellates, and cryptophytes (represented by letters, i.e., A-cyanobacteria, B-green algae, C-diatoms, D-dinoflagellates, E-cryptophytes) and then dilute them at a ratio of 1:1.5 to form 9 concentration gradients.

[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 experimental table is as follows:

[0158]

[0159]

[0160] Each experimental group was prepared by taking a certain volume of algal solution of the same genus and designing it according to the above orthogonal table. The final mixed algal solution was obtained by cross-mixing the volumes in the preferred volume ratio A:B:C:D:E = 1:1:1:1:1.

[0161] The above 81 groups of mixed algal solutions were subjected to parameter determination in step (3), including turbidity (represented by e), chlorophyll a (represented by g), and microcystin LR. The concentration of chlorophyll a in the samples was measured to be 0-600 μg / L, the turbidity to be 0-100 NTU, and the microcystin LR to be 0.5-15 μg / L.

[0162] The colored soluble organic compound (CDOM) (represented by h) prepared in step (3) was added to the mixed algal solution prepared above, and the concentration of CDOM in each sample was measured to be 0.6-10 mg / L.

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

[0164] According to the data matrix selection and optimization scheme in step (501), the preferred data matrix is ​​extracted and transformed into a one-dimensional array (i.e., for the data matrix M) g Data from excitation wavelengths of 351-660nm and emission wavelengths of 661-760nm are extracted and formed into a matrix, which is then converted into a one-dimensional array, denoted as M. y1 Its length is 6200; additionally, data from excitation wavelengths of 541-620nm and emission wavelengths of 622-661nm are extracted to form a matrix, which is converted into a one-dimensional array denoted as M. y2 Its length is 640; the one-dimensional array M y1 and M y2 Merge them 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 compiled into the following matrix:

[0167]

[0168] The microcystin LR values ​​y of each sample were used to construct a matrix Y. 81 The matrix size is 81×1;

[0169] Let matrix X 81 Sum matrix W 81 Form a new matrix M XW81 For matrix M XW81 Principal component analysis was performed to obtain the dimension-reduced matrix M. JXW81 Then matrix M JXW81 And matrix Y 81 Form a new matrix MJXWY81 , where Y 81 Located in matrix M JXWY81 The last column, for matrix M JXWY81 After normalization, matrix M is obtained. Jsdata81 ;

[0170]

[0171] M Jsdata81 The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge The model provides a predicted value for microcystin LR in each sample. The goodness of fit R of the regression model 2 >0.95.

[0172] Example 3

[0173] Five algal strains purchased from the Freshwater Algae Culture Bank of the Chinese Academy of Sciences were cultured according to step 1.

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

[0175] Category Algal species name Latin name coding Cyanobacteria Microcystis aeruginosa Microcystis aeruginosa AA Chlorophyta oblique tetra-chain algae Tetradesmus obliquus BB Diatoms Meniereella Cyclotella meneghiniana CC Dinophyta Shield-shaped Dinoflagellate Peridinium umbonatum var.inaequale DD Cryptophyta Cryptophytes Cryptomonas sp. EE

[0176] Fluorescence spectra of typical algal species and sodium humate are shown in [reference needed]. Figure 1-6 .

[0177] The culture media for the above algal species were BG11 medium for cyanobacteria and green algae, CSI medium for diatoms, 119 medium for dinoflagellates, and AF6 medium for cryptophytes; the culture media for each algal species were purchased from the Freshwater Algae Culture Bank of the Chinese Academy of Sciences.

[0178] The above culture was carried out to achieve the desired algal density of 10. 7 Five algal species with a cell / L or higher were mixed according to step (2) to obtain 256 sets of mixed algal solutions.

[0179] The microcystin LR of each sample of the above-mixed algal solution was determined according to step (2);

[0180] The above-mentioned mixed algal solution samples were analyzed according to step (2) to determine the parameters of each sample, including pH, conductivity, dissolved oxygen, water temperature, turbidity, algal density, chlorophyll a, and colored dissolved organic matter (CDOM). These parameters were represented by the symbols a, b, c, d, e, f, g, and h, respectively. The results of these parameter measurements were represented by the following matrix:

[0181]

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

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

[0184] Let matrix X 256 Sum matrix W 256 Form a new matrix M XW256 For matrix M XW256 Principal component analysis was performed to obtain the dimension-reduced matrix M. JXW256 Then matrix M JXW256 And matrix Y 256 Form a new matrix M JXWY256 , where Y 256 Located in matrix M JXWY256 The last column, for matrix M JXWY256 After normalization, matrix M is obtained. Jsdata256 ;

[0185]

[0186] M Jsdata256 The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge The model provides a predicted value for microcystin LR in each sample. The goodness of fit R of the regression model 2 >0.93.

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

[0188] Following step (503), substitute the above water sample measurement data 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 indicates not detected;

[0192] LR calculation value before optimization: Ridge regression modeling was performed solely based on scanned fluorescence spectral data.

[0193] Optimized LR calculation values: Ridge regression modeling was performed using optimized fluorescence spectral data and influencing factors such as dissolved oxygen, water temperature, turbidity, algal density, chlorophyll a, and colored dissolved organic matter (CDOM).

Claims

1. A method for retrieving microcystin content in water using algal fluorescence spectroscopy, characterized in that, Includes the following steps: (B1) Three-dimensional fluorescence scanning was performed on N algal samples with different concentrations. The excitation wavelength range was 351-660 nm, and the emission wavelength range was 451-800 nm. Data from the excitation wavelength range of 351-660 nm and the emission wavelength range of 661-760 nm were selected to form a matrix, which was then converted into a one-dimensional array M. y1 Furthermore, data from excitation wavelengths of 541-620nm and emission wavelengths of 622-661nm were extracted to form a matrix, which was then converted into a one-dimensional array M. y2 ; Transform the one-dimensional array M y1 and M y2 Merge into a one-dimensional array M p The one-dimensional arrays corresponding to N algal liquid samples are combined to form a matrix X. N Matrix X N Each row represents an independent variable for an algal solution sample, matrix X N The influence factors of N×p algal liquid samples are collected simultaneously and plotted in a matrix. Representation, matrix The size is N×m, where m is the number of influencing factors; (B2) The microcystin content of N algal solution samples with different concentrations was determined, and matrix Y was constructed. N The matrix size is N×1; (B3) Transform matrix X N sum matrix Composition matrix M XW Matrix M XW For a matrix M of size N×(p+m), XW Principal component analysis was performed to obtain the dimension-reduced matrix M. JXW Then matrix M JXW And matrix Y N Composition matrix M JXWY , where Y N Located in matrix M JXWY The last column, for matrix M JXWY After normalization, matrix M is obtained. Jsdata M Jsdata The last column in the algorithm is denoted as Y_Train, and the rest are denoted as X_Train. Substituting these into RidgeCV, cross-validation is used to automatically select the optimal regularization parameters to build the model. The resulting model is denoted as M. ridge ; (B4) Following the processing method in step (B1), the fluorescence spectrum data of the sample to be tested is measured, and after preprocessing, it is converted into a one-dimensional array M of length p. px Simultaneously, the influence factors of the sample to be tested are determined, and a one-dimensional array M is used. px The transformed matrix and the matrix composed of the influence factors are merged into a matrix Xp, and Xp is set as X_Test, and the LR prediction value of microcystin is Y_Pred. This value is then substituted into the established model M. ridge In the model, the predicted value Y_Pred is automatically provided; The influencing factors include one or more of the following: light intensity, water temperature, salinity, total alkalinity, turbidity, pH value, colored dissolved organic matter, and dissolved oxygen. The preprocessing includes blank removal, Raman normalization, Rayleigh and Raman scattering removal, data smoothing, and data matrix selection and optimization, specifically including the following steps: (101) Subtract the difference between the fluorescence spectral data matrix of the algal solution sample or the sample to be tested and the fluorescence spectral data matrix of the ultrapure water; (102) From the obtained Raman wavelength scanning data, the fluorescence data in the emission wavelength range of ultrapure water (371-421 nm) is extracted, integrated, and the Raman integral of the ultrapure water Raman peak is obtained, denoted as . Divide the fluorescence spectral data matrix of the algal solution sample or the sample to be tested by the Raman integral after subtracting the blank. The new data obtained are Raman unitized fluorescence spectra of algal liquid samples or samples to be tested; (103) The influence of Rayleigh scattering was removed by setting the fluorescence intensity to 0 within the range of ±20-50nm of the excitation wavelength; Raman scattering is an inelastic scattering of water molecules, and its influence was eliminated by subtracting the measured three-dimensional fluorescence spectrum of ultrapure water; the data were interpolated using the Delaunay triangle interpolation method to further remove or reduce the influence of Rayleigh scattering or Raman scattering. (104) The data obtained in step (103) is smoothed using a Gaussian kernel to reduce noise interference; (104) Convert the data matrix obtained in step (104) into a one-dimensional array M. p , or M px .

2. The method for retrieving microcystin content in water using algal fluorescence spectroscopy according to claim 1, characterized in that, The algal solution sample in step (B1) and the sample to be tested in step (B4) both include single algal solution samples and mixed algal solution samples. Furthermore, steps (A4) and (B4) should correspond consistently. Single algal solution preparation refers to taking a certain volume of algal seed and culturing it to reach the required algal density of 10-1. 7 The preparation of mixed algal solutions involves taking a certain volume of algal solutions from a specific phylum and genus of algae with a density of 10⁻⁶ cells / L or higher, and diluting them according to a specific dilution gradient to obtain a series of single algal solutions of different concentrations; the preparation of mixed algal solution samples refers to taking a certain volume of different genera of the same algal phylum that have reached the desired algal density of 10⁻⁶ cells / L. 7 Algae with a cell / L or higher concentration are mixed or diluted in a certain proportion to obtain a series of mixed algal solutions of various concentrations; or they can refer to a series of mixed algal solutions of various concentrations obtained by mixing or diluting algae of different phyla and genera in a certain proportion to obtain a certain volume of algae.

3. The method for retrieving microcystin content in water using algal fluorescence spectroscopy according to claim 1, characterized in that, In step (B2), the microcystin content was determined according to GB / T 20466-2006 "Determination of Microcystin in Water".

Images