A membrane fouling in-situ contaminant detection method based on front surface fluorescence data

Abstract

The application relates to the technical field of membrane pollution detection, in particular to a membrane pollution in-situ pollutant detection method based on front surface fluorescence data. The method adopts a machine learning model to perform nonlinear modeling on original fluorescence data containing multiple interferences, further introduces a symbolic regression method to analyze and process the prediction result of the machine learning model, obtains a membrane pollution in-situ pollutant relative concentration calculation formula, and corrects the relative concentration of a target pollutant, so that the collection of multiple fluorescence parameters and the correction of complex fluorescence data are not needed, the method has the comprehensive advantages of high prediction accuracy, few parameter dependencies and strong interference scene adaptability, and is suitable for in-situ fluorescence analysis of a membrane pollution process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of membrane fouling detection, and particularly relates to a membrane fouling in-situ pollutant detection method based on front surface fluorescence data. BACKGROUND

[0002] Three-dimensional fluorescence analysis method has the characteristics of high efficiency, sensitivity, rich information and no additional pretreatment, and is widely used in recent years to characterize natural organic matter with fluorescence characteristics in water. The front surface fluorescence technology is to make the excitation light beam form an incident angle of 30°-60° with the sample surface, so as to reduce the interference of light absorption and scattering on fluorescence detection. In the process of three-dimensional fluorescence detection, when the solution concentration satisfies UV254 less than 0.05, the absorbance of the solution to the excitation light is very low, and the fluorescence intensity of the measured solution is proportional to the concentration of the fluorescent substance. With the further increase of the solution concentration, the linear relationship between the fluorescence intensity and the solution concentration gradually fails. When the solution concentration is too high, due to the influence of internal filter effect, emission light reabsorption phenomenon and light scattering, the fluorescence intensity decreases with the increase of the concentration, which leads to the distortion phenomenon.

[0003] Due to the interference factors such as internal filter effect, emission light reabsorption phenomenon and light scattering, the front surface fluorescence data needs to be further corrected. At present, the methods for correcting fluorescence data mainly include three categories: one is the physical model correction based on optical principles, such as Kubelka-MonK theoretical correction model, which corrects the fluorescence data by theoretical function to calculate the total release coefficient, wherein is the total release coefficient, is the diffuse reflection coefficient at the corresponding wavelength , and then the correction coefficient is calculated by the correction function , wherein, is the correction function, is the incident wavelength, To excite the wavelength, finally divide the original fluorescence data by the correction factor to obtain the corrected fluorescence value. But this kind of method needs to add additional equipment to obtain reflectivity data, in high concentration or complex composition sample, using a single correction factor often difficult to accurately eliminate the error caused by internal filter effect, when the type or concentration of light absorbing substance in the sample changes, the correction factor needs to be re-determined and corrected. The second commonly used method is based on chemometrics or mathematical linear regression model, such as principal component regression, partial least squares regression, this kind of method has limited modeling ability for nonlinear system, the calibration set used for modeling needs to fully cover the variation range of all response components in the sample to be measured, otherwise the model generalization ability is insufficient. The third commonly used method is a modeling and correction method based on machine learning, which does not depend on physical optical model and does not need to measure absorbance or reflectivity additionally, has strong nonlinear modeling ability, but traditional machine learning model belongs to "black box" model, its internal decision process is difficult to analyze, lacks clear mathematical expression form, which is not conducive to mechanism research, knowledge transfer and practical engineering promotion. SUMMARY

[0004] The purpose of the present application is to provide a membrane pollution in-situ pollutant detection method based on front surface fluorescence data, which solves the above technical problems.

[0005] To achieve the above purpose, the present application provides a membrane pollution in-situ pollutant detection method based on front surface fluorescence data, the specific steps are as follows:

[0006] Step S1: obtaining front surface fluorescence data set before membrane pollution and pretreatment;

[0007] Step S2: extracting feature input value of front surface fluorescence data, the feature input value includes target pollutant characteristic fluorescence intensity and at least one Rayleigh scattering intensity related to target pollutant peak value;

[0008] Step S3: constructing fluorescence feature data set according to the extracted feature input value and known relative concentration value of membrane pollution, and training at least one machine learning model through the fluorescence feature data set and optimizing the machine learning model, and selecting the optimal machine learning model through model evaluation index;

[0009] Step S4: constructing symbolic regression model according to the prediction output value of the optimal machine learning model and the corresponding feature input value, and obtaining membrane pollution in-situ pollutant relative concentration calculation formula according to the symbolic regression model, the membrane pollution in-situ pollutant relative concentration calculation formula is as follows:

[0010]

[0011] Wherein, Y is the relative concentration of membrane pollution in-situ pollutant, is a correction parameter, the fluorescence intensity of the target pollutant is detected, the scattering intensity at the excitation wavelength of the target pollutant is detected;

[0012] Step S5: Collecting the surface fluorescence data before membrane fouling, and calculating the relative concentration of in-situ pollutants of membrane fouling according to the in-situ pollutant relative concentration calculation formula of membrane fouling.

[0013] Preferably, in step S1, the process of obtaining the surface fluorescence data set before membrane fouling is as follows:

[0014] Step S11: Configuring target pollutant solutions with different relative concentrations, and adding different kinds of interferents in the target pollutant solutions;

[0015] Step S12: Filtering the target pollutant solutions containing different interferents by using ultrafiltration membranes, and collecting the three-dimensional fluorescence spectrum data of the surface of the ultrafiltration membranes in a light-proof environment by using an optical fiber probe;

[0016] Step S13: Preprocessing the three-dimensional fluorescence spectrum data, which includes preprocessing of removing outliers, standardization or normalization.

[0017] Preferably, the target pollutant solution is algae solution, and the interferent is one or any combination of nanoscale silicon dioxide, humic acid, aluminum chloride, ferric sulfate, organic pigment red, organic pigment green, and organic pigment purple.

[0018] Preferably, the fluorescence spectrum detection range is excitation wavelength 610-660 nm and emission wavelength 610-660 nm, the fluorescence spectrum detection range is excitation wavelength 610-660 nm and emission wavelength 610-660 nm, the excitation light data interval is 2-10 nm, the emission light data interval is 0.5-5 nm, the excitation light slit width is 5-20 nm, the emission light slit width is 5-20 nm, and the scanning speed is 3000-60000 nm / min.

[0019] Preferably, the ultrafiltration membrane is a polyvinylidene fluoride flat plate ultrafiltration membrane.

[0020] Preferably, in step S3, the machine learning model is one or any combination of an extreme gradient boosting tree model, a random forest model, a stacked ensemble model, a multilayer perceptron-artificial neural network model, a K-nearest neighbor model, or a support vector regression model.

[0021] Preferably, in step S3, the optimization of the machine learning model includes:

[0022] optimizing the tree number, tree depth, learning rate, sample sampling ratio, feature sampling ratio, minimum split gain, and regularization parameter of the extreme gradient boosting tree model;

[0023] Cross-validation set is integrated to the stacked ensemble model through the meta-model;

[0024] The number of layers, the number of neurons, the learning rate and the regularization parameter are optimized for the multi-layer perception-artificial neural network model;

[0025] The K value, distance measure and weighted voting parameter are optimized for the K-nearest neighbor model;

[0026] The kernel function, regularization coefficient and tolerance range parameter are optimized for the support vector regression model in combination with grid search and cross-validation.

[0027] Preferably, the model evaluation indicators include goodness of fit, mean square deviation, mean absolute deviation, root mean square error and mean relative error.

[0028] Preferably, the membrane pollution in-situ pollutant relative concentration calculation formula is selected from the symbolic regression model output according to the algae absorption interference compensation mechanism and the fitting accuracy.

[0029] Therefore, the membrane pollution in-situ pollutant detection method based on the front surface fluorescence data has the beneficial effects that:

[0030] (1) By introducing the machine learning model, the complex nonlinear relationship between the characteristic fluorescence peak of the target pollutant in the original three-dimensional fluorescence signal and the interference such as Rayleigh scattering in the adjacent region is modeled, and a high-precision relative concentration prediction result is obtained. The high-precision relative concentration prediction result is introduced into the symbolic regression model, the analytical function relationship between the input features and the output results is excavated, without additional measurement of optical parameters such as absorbance and reflectivity, and without complex correction of fluorescence data, only relying on the original three-dimensional fluorescence data can complete the data correction of membrane pollution in-situ pollutants.

[0031] (2) In combination with the characteristic fluorescence intensity of the specific target pollutant and the Rayleigh scattering signal in the neighborhood as the key feature value input, and adding different types of interference substances in the training data set, the model's ability to extract effective information under complex interference background is enhanced, thereby efficiently completing the algae density modeling and interference correction task, and improving the robustness, generalization ability and actual prediction accuracy of the model.

[0032] The technical solutions of the present application will be further described in detail below with the aid of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 A flowchart of the membrane pollution in-situ pollutant detection method based on the front surface fluorescence data of the present application;

[0034] Figure 2Fitting goodness histogram for the case of using pure algae solution without adding interference in the first group;

[0035] Figure 3 Fitting goodness histogram for the case of using algae solution with added interference in the second group; DETAILED DESCRIPTION

[0036] In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer" and the like indicate the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly used when the product of the present application is used, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "arrangement", "installation", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0037] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0038] As Figure 1 shown, a membrane pollution in-situ pollutant detection method based on front surface fluorescence data, the specific steps are as follows:

[0039] Step S1: Obtain the front surface fluorescence data set before membrane pollution and pre-process.

[0040] In step S1, the process of obtaining the front surface fluorescence data set before membrane pollution is as follows:

[0041] Step S11: Configure target pollutant solutions with different relative concentrations, and add different kinds of interference in the target pollutant solutions. The target pollutant solution of the present embodiment is algae solution, and pure algae solution is configured from raw water of Microcystis aeruginosa from Freshwater Algae Culture Collection of Institute of Hydrobiology, Chinese Academy of Sciences, and the interference is one or any combination of nano-sized silicon dioxide, humic acid, aluminum chloride, ferric sulfate, organic pigment red, organic pigment green and organic pigment purple.

[0042] Step S12: Filter the target pollutant solution containing different interference by using an ultrafiltration membrane. The ultrafiltration membrane is a polyvinylidene fluoride flat ultrafiltration membrane, and the polyvinylidene fluoride flat ultrafiltration membrane produced by Guochu Science and Technology (Xiamen) Co., Ltd. is used. The effective filtration area of the membrane is 45cm 2After each filtration volume of algal liquid, the three-dimensional fluorescence spectrum data of the front surface of the ultrafiltration membrane was collected by the optical fiber probe in a light-proof environment; the fluorescence spectrum detection range was excitation wavelength 610-660 nm, emission wavelength 610-660 nm, scanning interval was excitation wavelength 5 nm, emission wavelength 2 nm, slit width was 10 nm, and scanning speed was 30000 nm / s.

[0043] Step S13: data preprocessing was performed on the three-dimensional fluorescence spectrum data, the data preprocessing included removing outliers, standardization processing or normalization preprocessing, the three-dimensional fluorescence spectrum data derived txt format file was converted into excel format convenient for reading and identification, and related preprocessing was performed according to actual subsequent model needs.

[0044] Step S2: the characteristic input value of the front surface fluorescence data was extracted, the characteristic input value included target pollutant characteristic fluorescence intensity and two-place labeled pollutant peak related Rayleigh scattering intensity. The two-place labeled pollutant peak related Rayleigh scattering intensity was the Rayleigh scattering fluorescence intensity at excitation wavelength / emission wavelength 610 nm / 610 nm and the Microcystis aeruginosa characteristic fluorescence intensity at 610 nm / 660 nm extracted from the fluorescence spectrum data as the characteristic input value.

[0045] The real algal cell density was calculated according to the existing formula relationship between the cell concentration of Microcystis aeruginosa and the absorbance, and the cell concentration of the algal liquid was calculated, and then the algal cell density was calculated according to the membrane area. The real algal cell density was calculated according to the existing formula relationship between the cell concentration of Microcystis aeruginosa and the absorbance, and the cell concentration of the algal liquid was calculated, and then the algal cell density was calculated according to the membrane area.

[0046] The input value and the output value of the data set were labeled and classified, the Rayleigh scattering fluorescence intensity at excitation wavelength / emission wavelength 610 nm / 610 nm in the input value was x1, the Microcystis aeruginosa characteristic fluorescence intensity at excitation wavelength / emission wavelength 610 nm / 660 nm was x2, the Rayleigh scattering fluorescence intensity at excitation wavelength / emission wavelength 660 nm / 660 nm was x3, and the real algal cell density was y.

[0047] Step S3: a fluorescence characteristic data set was constructed according to the extracted characteristic input value and the known membrane pollution relative concentration value, the fluorescence characteristic data set was divided into a training set and a test set according to a ratio of 8:2, six machine learning models were trained and tested through the fluorescence characteristic data set, the machine learning models were optimized, and the optimal machine learning model was selected through model evaluation indexes.

[0048] The six machine learning models are an extreme gradient boosting tree model, a random forest model, a stacked ensemble model, a multilayer perceptron-artificial neural network model, a K-nearest neighbor model, and a support vector regression model. The tree number, tree depth, learning rate, sample sampling ratio, feature sampling ratio, minimum split gain, and regularization parameter of the extreme gradient boosting tree model are optimized; the stacked ensemble model is integrated by cross-validation of the meta-model; the number of layers, the number of neurons, the learning rate, and the regularization parameter of the multilayer perceptron-artificial neural network model are optimized; the K value, the distance metric, and the weighted voting parameter of the K-nearest neighbor model are optimized; the kernel function, the regularization coefficient, and the tolerance range parameter of the support vector regression model are optimized in combination with grid search and cross-validation to adjust the parameters and control the generalization error.

[0049] The model evaluation indicators include goodness of fit R 2 , mean square error MSE, mean absolute error MAE, root mean square error RMSE, and mean relative error MeanRE.

[0050] Step S4: According to the prediction result data set of the optimal machine learning model and the corresponding feature input value, a membrane pollution in-situ pollutant relative concentration calculation formula is constructed by inputting the constructed symbolic regression modeling. The above derived data is modeled and analyzed using the SymbolicRegression.jl symbolic regression tool, and the membrane pollution in-situ pollutant relative concentration calculation formula is selected from the symbolic regression model output according to the algae absorption interference compensation mechanism and the fitting accuracy.

[0051] The membrane pollution in-situ pollutant relative concentration calculation formula is as follows:

[0052]

[0053] Wherein, Y is the membrane pollution in-situ pollutant relative concentration, is a correction parameter, is the fluorescence intensity of the target pollutant, is the scattering intensity at the excitation wavelength of the target pollutant;

[0054] Step S5: Collect the surface fluorescence data before membrane pollution, and calculate the relative concentration of the membrane pollution in-situ pollutant according to the membrane pollution in-situ pollutant relative concentration calculation formula.

[0055] In order to verify the anti-interference ability of the technical scheme of the embodiment, a control test is carried out:

[0056] The first group adopts pure algal liquid without adding interference, the second group adopts algal liquid with added interference, and no correction method, Kubelka-MonK theory correction model and the technical means of the embodiment are used respectively. 、 The total slow-release coefficient is calculated by a theoretical function , and the correction coefficient is calculated by a correction function . Finally, the original fluorescence data is divided by the correction coefficient to obtain the fluorescence value after KM theory correction, which is fitted with the algal cell density.

[0057] The goodness of fit is shown in Figures 2-3 . It can be seen that in the pure algal liquid, the goodness of fit of the three technical means is similar, but the performance increases in turn. In the presence of interference (the interference is a combination of nano-sized silicon dioxide, humic acid, aluminum chloride, ferric sulfate, organic pigment red, organic pigment green and organic pigment purple), the goodness of fit of no correction method and Kubelka-MonK theory correction model decreases to 0.507 and 0.534 respectively, while the goodness of fit of the technical solution of the embodiment is 0.8847, which is high in reliability and strong in interference scene adaptation.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can still be modified or replaced by equivalents, and these modifications or replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for in-situ detection of membrane contaminants based on front surface fluorescence data, characterized in that, The specific steps are as follows: Step S1: Obtain the surface fluorescence dataset before membrane fouling and perform preprocessing; Step S2: Extract the feature input values ​​of the front surface fluorescence data. The feature input values ​​include the characteristic fluorescence intensity of the target pollutant and at least one Rayleigh scattering intensity related to the peak value of the target pollutant. Step S3: Construct a fluorescence feature dataset based on the extracted feature input values ​​and the known relative concentration values ​​of membrane fouling, and train at least one machine learning model using the fluorescence feature dataset and optimize the machine learning model. Select the optimal machine learning model using model evaluation metrics. Step S4: Construct a symbolic regression model based on the predicted output of the optimal machine learning model and the corresponding feature input values. Then, derive the formula for calculating the relative concentration of in-situ pollutants in membrane fouling based on the symbolic regression model. The formula for calculating the relative concentration of in-situ pollutants in membrane fouling is as follows: Where Y represents the relative concentration of in-situ pollutants in membrane fouling. To correct the parameters, To detect the fluorescence intensity of the target pollutant, The scattering intensity at the excitation wavelength of the target pollutant; Step S5: Collect surface fluorescence data before membrane fouling, and calculate the relative concentration of in-situ pollutants in membrane fouling according to the formula for calculating the relative concentration of in-situ pollutants in membrane fouling.

2. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 1, characterized in that: In step S1, the process of obtaining the surface fluorescence dataset before membrane fouling is as follows: Step S11: Prepare target pollutant solutions with different relative concentrations and add different types of interfering substances to the target pollutant solutions; Step S12: Use an ultrafiltration membrane to filter a solution of target pollutants containing different interfering substances, and collect three-dimensional fluorescence spectral data of the front surface of the ultrafiltration membrane using a fiber optic probe in a light-protected environment; Step S13: Perform data preprocessing on the three-dimensional fluorescence spectroscopy data. Data preprocessing includes outlier removal, standardization, or normalization.

3. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 2, characterized in that: The target pollutant solution is algal solution, and the interfering substances are one or any combination of nano-sized silica, humic acid, aluminum chloride, ferric sulfate, organic pigment red, organic pigment green, and organic pigment purple.

4. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 3, characterized in that: The fluorescence spectroscopy detection range is 610nm-660nm for excitation wavelength and 610nm-660nm for emission wavelength. The excitation data interval is 2-10nm, the emission data interval is 0.5-5nm, the excitation slit width is 5-20nm, the emission slit width is 5-20nm, and the scanning speed is 3000-60000nm / min.

5. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 4, characterized in that: The ultrafiltration membrane is a polyvinylidene fluoride flat sheet ultrafiltration membrane.

6. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 5, characterized in that: In step S3, the machine learning model is one or any combination of extreme gradient boosting tree model, random forest model, stacked ensemble model, multilayer perceptron-artificial neural network model, K-nearest neighbor model or support vector regression model.

7. The in-situ contaminant detection method for membrane fouling based on front surface fluorescence data according to claim 6, characterized in that: In step S3, the optimization of the machine learning model includes: Optimize the number of trees, tree depth, learning rate, sample sampling ratio, feature sampling ratio, minimum split gain, and regularization parameters for extreme gradient boosting tree models; The stacked ensemble model is cross-validated and integrated using a meta-model. Optimize the number of layers, number of neurons, learning rate, and regularization parameters for the multilayer perceptron-artificial neural network model; The K-nearest neighbor model optimizes the K value, distance metric, and weighted voting parameters. The kernel function, regularization coefficient, and tolerance range parameters of the support vector regression model are optimized by combining grid search and cross-validation for parameter tuning.

8. The method for in-situ detection of membrane contaminants based on front surface fluorescence data according to claim 7, characterized in that: Model evaluation metrics include goodness of fit, mean squared deviation, mean absolute deviation, root mean square error, and mean relative error.

9. The in-situ contaminant detection method for membrane fouling based on front surface fluorescence data according to claim 8, characterized in that: The formula for calculating the relative concentration of in-situ pollutants in membrane fouling was selected from the output of the symbolic regression model based on the algal absorption interference compensation mechanism and fitting accuracy.

Images