Prediction method for antibiotic removal effect in water based on synergistic effect of plants and activated carbon

Abstract

The present invention discloses a method for predicting the antibiotic removal effect in water bodies based on the synergy of plants and activated carbon. The method comprises the following steps: obtaining historical data and data to be predicted; performing similarity matching on the historical data based on the data to be predicted to obtain similar small sample data; dividing the similar small sample data into stages based on the rate of change of antibiotic concentration in the water body to obtain first-stage data and second-stage data; determining a rapid adsorption range and first adsorption data based on the first-stage data and the data to be predicted; generating extended data based on the second-stage data, and screening the extended data for high-confidence samples based on the rapid adsorption range; constructing a charcoal plant adsorption model based on the high-confidence samples and performing hyperparameter optimization using the Longying Oat Optimization Algorithm (AOO); and inputting the first adsorption data into the charcoal plant adsorption model to output a prediction result. This method improves the accuracy of the prediction of the antibiotic removal effect in water bodies based on the synergy of plants and activated carbon and has strong interpretability.

Description

Technical Field

[0001] The present invention relates to the field of water body antibiotic removal effect prediction, and in particular to a water body antibiotic removal effect prediction method based on the synergy of plants and activated carbon. Background Art

[0002] With the widespread use of antibiotics in medicine, animal husbandry, and aquaculture, the problem of antibiotic residues in water bodies is becoming increasingly serious. These pollutants not only disrupt aquatic ecosystems but can also accumulate through the food chain, threatening human health.

[0003] Traditional methods for predicting the removal effect of antibiotics in water bodies mainly rely on large-scale historical data of single-action plants or activated carbon to predict the removal effect, ignoring the fact that removal under synergistic action will have a phased characteristic. In addition, the prediction method for the removal effect of antibiotics in water bodies using the synergistic effect of plants and activated carbon often faces a shortage of small sample data in actual scenarios, making it difficult to capture the removal rules in complex environments. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Obtain historical data on the synergistic removal of antibiotics from water bodies by plants and activated carbon, as well as data to be predicted for water bodies to be predicted;

[0008] Performing similarity matching on historical data according to the data to be predicted to obtain similar small sample data, and dividing the similar small sample data into stages according to the rate of change of antibiotic concentration in water to obtain first stage data and second stage data;

[0009] Determining a rapid adsorption range and first adsorption data based on the first-stage data and the data to be predicted;

[0010] Generate extended data based on the second-stage data transfer learning, and screen high-confidence samples from the extended data in combination with the rapid adsorption range;

[0011] A charcoal plant adsorption model was constructed based on the high-confidence sample and hyperparameter optimization was performed using the AOO optimization algorithm of Longying Oats;

[0012] The first adsorption data is input into the carbon plant adsorption model to output a prediction result.

[0013] Furthermore, the acquisition of historical data on the synergistic removal of antibiotics from water bodies by plants and activated carbon and data to be predicted for the water body to be predicted includes:

[0014] The historical data include time series data consisting of water antibiotic concentration, water pH value, water temperature, plant height, and activated carbon input collected daily when plants and activated carbon synergistically remove water antibiotics in different water bodies;

[0015] The data to be predicted include water antibiotic concentration, water pH value, water temperature, activated carbon input and plant height.

[0016] Furthermore, the method of performing similarity matching on historical data according to the data to be predicted to obtain similar small sample data includes:

[0017] extracting initial condition data of the historical data;

[0018] Calculate the similarity between the initial condition data and the data to be predicted. The similarity calculation expression is as follows:

[0019]

[0020] in is the parameter vector of the data to be predicted and the initial condition data, is the initial condition data dimension, is the initial condition data The mean of the parameters, is the initial condition data The standard deviation of the parameters, The data to be predicted The value of the parameter, The initial condition data The value of the parameter;

[0021] According to the water body initial condition data whose similarity meets a preset similarity threshold, the corresponding collected time series data is extracted from the historical data as similar small sample data.

[0022] Furthermore, a method for obtaining first-stage data and second-stage data by dividing the similar small sample data into stages according to the rate of change of antibiotic concentration in water bodies includes:

[0023] Extract the antibiotic concentration time series of the water body of the similar small sample data according to the initial conditions;

[0024] Calculate the rate of decrease of antibiotic concentration in water between two adjacent time points;

[0025] The least squares method was used to perform a segmented linear regression analysis on the rate of decrease of the antibiotic concentration in the water body, and the stage transition points with significantly different slopes were identified through the F test and used as critical points;

[0026] According to the critical points under different conditions, all data in the similar small sample data from the initial moment to the critical point are divided into first-stage data, and the data after the critical point are divided into second-stage data;

[0027] Furthermore, the method for determining the rapid adsorption range and the first adsorption data based on the first stage data and the data to be predicted includes:

[0028] Extract the data corresponding to the last time under different initial conditions according to the first stage data as the first stage effect data;

[0029] The maximum removal rate and the minimum removal rate of the antibiotic concentration in the water body in the effect data of the first stage are used as the rapid adsorption range;

[0030] The antibiotic concentration time series of the water body under different initial conditions in the first stage data is fitted by the LM algorithm to obtain an antibiotic concentration index, and then the first concentration data is obtained by weighting according to the similarity between the initial condition data and the data to be predicted;

[0031] The first feature data is obtained by weighting the difference between the effect data of the first stage and the initial condition data by similarity;

[0032] The first characteristic data and the first concentration data are spliced ​​according to the dimension of the data to be predicted to obtain the first adsorption data.

[0033] Furthermore, based on the second-stage data transfer learning to generate extended data, a method for screening high-confidence samples from the extended data in combination with the rapid adsorption range includes:

[0034] The temporal convolutional network (TCN) is used to extract the dependent features of the water antibiotic concentration, water pH value, water temperature and plant height in the historical data to generate a temporal feature matrix. , where m is the initial number of water bodies, t is the length of the time series, and d is the data dimension. The time series feature matrix is ​​mean pooled to convert each time series feature into a statistical feature vector of fixed dimension. ;

[0035] The singular value decomposition method is used to reduce the dimension of the statistical characteristic matrix formed by the statistical characteristic vectors, and the first k principal components are retained as subspace basis vectors based on the cumulative variance contribution rate greater than 95%. ;

[0036] Calculate the statistical eigenvector With the subspace basis vector The feature importance weight of is calculated as follows: ,in is the inner product operation;

[0037] The extended feature vector is generated based on the statistical feature vector, subspace basis vector and feature importance weight. The calculation formula is as follows:

[0038]

[0039] in, is the extended eigenvector, is the inner product operation, The target feature vector to be expanded is composed of the data to be predicted. are the subspace basis vectors, is the feature importance weight, is the statistical eigenvector, is a random perturbation vector that satisfies physical constraints;

[0040] The maximum mean difference is used to evaluate the distribution difference between the extended feature vector and the original historical data. If the maximum mean difference is less than the preset difference threshold, the extended feature vector is retained;

[0041] Repeatedly generating multiple extended feature vectors to form extended data;

[0042] The initial antibiotic concentration of the extended data is extracted, and the antibiotic removal rate of the water body under different initial conditions of the extended data is calculated based on the initial antibiotic concentration. The samples with removal rates outside the rapid adsorption range are eliminated, and the antibiotic concentration of the initial data is tested according to the 3σ principle to screen out high-confidence samples.

[0043] Furthermore, a method for constructing a charcoal plant adsorption model based on the high-confidence sample and performing hyperparameter optimization using the long-winged oat optimization algorithm AOO includes:

[0044] A GRU neural network model was constructed, using the root mean square error as the fitness function. The input layer dimension was the sum of the dimension of the data to be predicted and the dimension of the time parameter. It contained two layers of GRU units, the activation function was a linear function, and the output was the predicted value of the antibiotic concentration in the water body.

[0045] AOO initialization uses Tent chaos mapping to initialize the population, and each individual is a combination of the number of hidden layer neurons, learning rate, Dropout ratio and weight matrix initialization;

[0046] Generate a random number between 0 and 1 ,when The update formula of oat position in the exploration phase is:

[0047]

[0048] in and The tth and The position of the i-th oat individual in generation, is the population size, is the dynamic adjustment factor, is the maximum number of iterations, is a random vector of 0 to 1 with the same dimension dim among the individuals in the population, It is Hadamard. is the upper bound vector of the target problem, is the position of the best individual in the group;

[0049] otherwise The update formula of oat position in the development stage is:

[0050] otherwise The update formula of oat position in the development stage is:

[0051]

[0052] in Indicates the quality of oat seeds, Represents the operation of independently generating random numbers from 0 to 1, is a randomly generated value between 0 and 1, dim represents the dimension of the target problem, Eccentricity coefficient during seed rolling, Indicates the length of the main awn of oat seeds, express, Indicates generating a random matrix with the same dimension as dim and values ​​between -A and A. , , dim is the dimension of the target problem, the characteristic exponent of the vegetable dimensional distribution is 1.5, which is used to control the distribution characteristics of the random step size, is the result of Levy flight function;

[0053] The algorithm is terminated when the maximum number of iterations is reached, and the optimal hyperparameter combination corresponding to the global optimal position is output;

[0054] The optimal hyperparameter combination is used to input a GRU neural network model, the extended data set is used for training, and the trained GRU neural network model is used as a carbon plant adsorption model.

[0055] The beneficial effects of the present invention are:

[0056] The present invention is a method for predicting the removal effect of antibiotics in water based on the synergy of plants and activated carbon. Compared with the existing technology, the present invention has the following technical effects:

[0057] The present invention can improve the accuracy of the prediction of the antibiotic removal effect of water bodies through data similarity matching, adsorption stage division, determination of the rapid adsorption range and the first adsorption data, data expansion and improvement of data reliability, acquisition model construction and optimization model steps, thereby improving the precision of the prediction of the antibiotic removal effect of water bodies, optimizing the prediction of the antibiotic removal effect of water bodies, greatly saving resources, improving work efficiency, and realizing intelligent prediction of the antibiotic removal effect of water bodies. It is of great significance to the prediction of the antibiotic removal effect of water bodies, can adapt to the prediction requirements of the antibiotic removal effect of water bodies with different standards, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of the steps of the method for predicting the removal effect of antibiotics in water based on the collaboration of plants and activated carbon of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0060] The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon of the present invention comprises the following steps:

[0061] like Figure 1 As shown, in this embodiment, the following steps are included:

[0062] Obtain historical data on the synergistic removal of antibiotics from water bodies by plants and activated carbon, as well as data to be predicted for water bodies to be predicted;

[0063] Performing similarity matching on historical data according to the data to be predicted to obtain similar small sample data, and dividing the similar small sample data into stages according to the rate of change of antibiotic concentration in water to obtain first stage data and second stage data;

[0064] Determining a rapid adsorption range and first adsorption data based on the first-stage data and the data to be predicted;

[0065] Generate extended data based on the second-stage data transfer learning, and screen high-confidence samples from the extended data in combination with the rapid adsorption range;

[0066] A charcoal plant adsorption model was constructed based on the high-confidence sample and hyperparameter optimization was performed using the AOO optimization algorithm of Longying Oats;

[0067] The first adsorption data is input into the carbon plant adsorption model to output a prediction result.

[0068] In this embodiment, the method for obtaining and preprocessing historical data of the synergistic removal of antibiotics from water bodies by plants and activated carbon and data to be predicted for the water body to be predicted includes:

[0069] The historical data include time series data consisting of water antibiotic concentration, water pH value, water temperature, plant height, and activated carbon input collected daily when plants and activated carbon synergistically remove water antibiotics in different water bodies;

[0070] The data to be predicted include water antibiotic concentration, water pH value, water temperature, activated carbon input and plant height;

[0071] In the actual assessment, the sampled water was stored at 4 degrees Celsius and the antibiotic concentration in the water was analyzed by high-performance liquid chromatography-tandem mass spectrometry HPLC-MS. The historical data included 20 sets of initial condition data and time series data collected under water bodies. Taking the predicted data of a certain water body as an example, its water antibiotic concentration was 120 ug / L, pH was 7.8, water temperature was 23.5 degrees Celsius, activated carbon input was 20 mg / L, and plant height was 26 cm.

[0072] In this embodiment, the method of performing similarity matching on historical data according to the data to be predicted to obtain similar small sample data includes:

[0073] extracting initial condition data of the historical data;

[0074] Calculate the similarity between the initial condition data and the data to be predicted. The similarity calculation expression is as follows:

[0075]

[0076] in is the parameter vector of the data to be predicted and the initial condition data, is the initial condition data dimension, is the initial condition data The mean of the parameters, is the initial condition data The standard deviation of the parameters, The data to be predicted The value of the parameter, The initial condition data The value of the parameter;

[0077] Extracting corresponding collected time series data from the historical data as similar small sample data based on the water body initial condition data whose similarity meets a preset similarity threshold;

[0078] In actual evaluation, the parameter vector of the initial condition data of a water body in historical data is taken as an example. [115, 7.5, 22.0, 18, 25], the parameter vector of the data to be predicted is [120, 7.8, 23.5, 20, 26]. The mean of water antibiotic concentration, water pH value, water temperature, activated carbon input and plant height in the historical data is [118, 7.7, 22.8, 20, 26], and the variance is [5.2, 0.2, 1.2, 1.8, 2.5]. The optimal similarity threshold is determined by cross-validation, and the value range is usually [0.6, 0.8]. In this actual evaluation, 0.7 is selected as the similarity threshold. The calculated similarity is 0.748, which meets the preset similarity threshold of 0.7. The data corresponding to the water body in the historical data is placed in the similar small sample data.

[0079] In this embodiment, the method for obtaining first-stage data and second-stage data by dividing the similar small sample data into stages according to the rate of change of antibiotic concentration in water bodies includes:

[0080] Extract the antibiotic concentration time series of the water body of the similar small sample data according to the initial conditions;

[0081] Calculate the rate of decrease of antibiotic concentration in water between two adjacent time points;

[0082] The least squares method was used to perform a segmented linear regression analysis on the rate of decrease of the antibiotic concentration in the water body, and the stage transition points with significantly different slopes were identified through the F test and used as critical points;

[0083] According to the critical points under different conditions, all data in the similar small sample data from the initial moment to the critical point are divided into first-stage data, and the data after the critical point are divided into second-stage data;

[0084] In the actual evaluation, taking the first five time points of the time series of antibiotic concentration in a certain water body under initial conditions after similarity matching in historical data as an example [115, 92, 70, 50, 45, 42], the F test determines that the critical point is 50, then the data of this water body at and before this point are classified as the first stage data, and the data of this point and after this point are classified as the second stage data.

[0085] In this embodiment, the method for determining the rapid adsorption range and the first adsorption data based on the first-stage data and the data to be predicted includes:

[0086] Extract the data corresponding to the last time under different initial conditions according to the first stage data as the first stage effect data;

[0087] The maximum removal rate and the minimum removal rate of the antibiotic concentration in the water body in the effect data of the first stage are used as the rapid adsorption range;

[0088] The antibiotic concentration time series of the water body under different initial conditions in the first stage data is fitted by the LM algorithm to obtain an antibiotic concentration index, and then the first concentration data is obtained by weighting according to the similarity between the initial condition data and the data to be predicted;

[0089] The first feature data is obtained by weighting the difference between the effect data of the first stage and the initial condition data by similarity;

[0090] splicing the first characteristic data and the first concentration data according to the dimension of the data to be predicted to obtain first adsorption data;

[0091] In the actual evaluation, the maximum removal efficiency was 56.5%, the minimum removal efficiency was 44.4%, and the first adsorption data were [55.9, 7.7, 23.4, 20, 28.4].

[0092] In this embodiment, the method of generating extended data based on the second-stage data transfer learning and screening high-confidence samples from the extended data in combination with the rapid adsorption range includes:

[0093] The temporal convolutional network (TCN) is used to extract the dependent features of the water antibiotic concentration, water pH value, water temperature and plant height in the historical data to generate a temporal feature matrix. , where m is the initial number of water bodies, t is the length of the time series, and d is the data dimension. The time series feature matrix is ​​mean pooled to convert each time series feature into a statistical feature vector of fixed dimension. ;

[0094] The singular value decomposition method is used to reduce the dimension of the statistical characteristic matrix formed by the statistical characteristic vectors, and the first k principal components are retained as subspace basis vectors based on the cumulative variance contribution rate greater than 95%. ;

[0095] Calculate the statistical eigenvector With the subspace basis vector The feature importance weight of is calculated as follows: ,in is the inner product operation;

[0096] The extended feature vector is generated based on the statistical feature vector, subspace basis vector and feature importance weight. The calculation formula is as follows:

[0097]

[0098] in, is the extended eigenvector, is the inner product operation, The target feature vector to be expanded is composed of the data to be predicted. are the subspace basis vectors, is the feature importance weight, is the statistical eigenvector, is a random perturbation vector that satisfies physical constraints;

[0099] The maximum mean difference is used to evaluate the distribution difference between the extended feature vector and the original historical data. If the maximum mean difference is less than the preset difference threshold, the extended feature vector is retained;

[0100] Repeatedly generating multiple extended feature vectors to form extended data;

[0101] Extracting the initial antibiotic concentration of the extended data, calculating the antibiotic removal rate of the water body under different initial conditions of the extended data based on the initial antibiotic concentration, eliminating samples with removal rates not within the rapid adsorption range, and screening out high-confidence samples after performing a 3σ principle test on the antibiotic concentration of the initial data;

[0102] In actual evaluation, the random perturbation vector The physical constraint is 0.1 times the variance of each dimension, and the initial concentration of the generated extended data is checked to see if it is within the range of the initial concentration of the historical data that satisfies the 3σ principle.

[0103] In this embodiment, a method for constructing a charcoal plant adsorption model based on the high-confidence sample and performing hyperparameter optimization using the long-winged oat optimization algorithm AOO includes:

[0104] A GRU neural network model was constructed, using the root mean square error as the fitness function. The input layer dimension was the sum of the dimension of the data to be predicted and the dimension of the time parameter. It contained two layers of GRU units, the activation function was a linear function, and the output was the predicted value of the antibiotic concentration in the water body.

[0105] AOO initialization uses Tent chaos mapping to initialize the population, and each individual is a combination of the number of hidden layer neurons, learning rate, Dropout ratio and weight matrix initialization;

[0106] Generate a random number between 0 and 1 ,when The update formula of oat position in the exploration phase is:

[0107]

[0108] in and The tth and The position of the i-th oat individual in generation, is the population size, is the dynamic adjustment factor, is the maximum number of iterations, is a random vector of 0 to 1 with the same dimension dim among the individuals in the population, It is Hadamard. is the upper bound vector of the target problem, is the position of the best individual in the group;

[0109] otherwise The update formula of oat position in the development stage is:

[0110]

[0111] in Indicates the quality of oat seeds, Represents the operation of independently generating random numbers from 0 to 1, is a randomly generated value between 0 and 1, dim represents the dimension of the target problem, Eccentricity coefficient during seed rolling, Indicates the length of the main awn of oat seeds, express, Indicates generating a random matrix with the same dimension as dim and values ​​between -A and A. , , dim is the dimension of the target problem, the characteristic exponent of the vegetable dimensional distribution is 1.5, which is used to control the distribution characteristics of the random step size, is the result of Levy flight function;

[0112] The algorithm is terminated when the maximum number of iterations is reached, and the optimal hyperparameter combination corresponding to the global optimal position is output;

[0113] Using the optimal hyperparameter combination to input the GRU neural network model, using the extended data set for training, and using the trained GRU neural network model as the charcoal adsorption model;

[0114] It should be explained that the input of the GRU neural network model is a feature vector of a single time step representing the end state of the first stage. The recommended maximum number of iterations in the AOO algorithm is 100.

[0115] In the actual evaluation, the first adsorption data [55.9, 7.7, 23.4, 20, 28.4] were input into the charcoal plant adsorption model to obtain a predicted value of 12.53 ug / L when the antibiotic concentration in the water body reached a stable state.

[0116] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the removal of antibiotics from water based on the synergy of plants and activated carbon, characterized in that: The following steps are involved: Obtain historical data on the synergistic removal of antibiotics from water bodies by plants and activated carbon, as well as data to be predicted for water bodies to be predicted; Performing similarity matching on historical data according to the data to be predicted to obtain similar small sample data, and dividing the similar small sample data into stages according to the rate of change of antibiotic concentration in water to obtain first stage data and second stage data; Determining a rapid adsorption range and first adsorption data based on the first-stage data and the data to be predicted; The process of determining the rapid adsorption range is as follows: Extract the data corresponding to the last time under different initial conditions according to the first stage data as the first stage effect data; The maximum removal rate and the minimum removal rate of the antibiotic concentration in the water body in the effect data of the first stage are used as the rapid adsorption range; The first adsorption data determination process is: The antibiotic concentration time series of the water body under different initial conditions in the first stage data is fitted by the LM algorithm to obtain an antibiotic concentration index, and then the first concentration data is obtained by weighting according to the similarity between the initial condition data of the historical data and the data to be predicted; The first feature data is obtained by weighting the difference between the effect data of the first stage and the initial condition data by similarity; splicing the first characteristic data and the first concentration data according to the dimension of the data to be predicted to obtain first adsorption data; Generate extended data based on the second-stage data transfer learning, and screen high-confidence samples from the extended data in combination with the rapid adsorption range; A charcoal plant adsorption model was constructed based on the high-confidence samples and hyperparameter optimization was performed using the AOO algorithm; The first adsorption data is input into the carbon plant adsorption model to output a prediction result.

2. The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon according to claim 1, characterized in that: The method for obtaining and preprocessing historical data of the collaborative removal of antibiotics from water bodies by plants and activated carbon and data to be predicted for water bodies to be predicted comprises: The historical data include time series data consisting of water antibiotic concentration, water pH value, water temperature, plant height, and activated carbon input collected daily when plants and activated carbon synergistically remove water antibiotics in different water bodies; The data to be predicted include water antibiotic concentration, water pH value, water temperature, activated carbon input and plant height.

3. The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon according to claim 1, characterized in that: The method for obtaining similar small sample data by performing similarity matching on historical data according to the data to be predicted includes: extracting initial condition data of the historical data; Calculate the similarity between the initial condition data and the data to be predicted. The similarity calculation expression is as follows: ； in is the parameter vector of the data to be predicted and the initial condition data, is the initial condition data dimension, is the initial condition data The mean of the parameters, is the initial condition data The standard deviation of the parameters, The data to be predicted The value of the parameter, The initial condition data The value of the parameter; According to the water body initial condition data whose similarity meets a preset similarity threshold, the corresponding collected time series data is extracted from the historical data as similar small sample data.

4. The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon according to claim 1, characterized in that: The method for obtaining first-stage data and second-stage data by dividing the similar small sample data into stages according to the change rate of antibiotic concentration in water bodies includes: Extract the antibiotic concentration time series of the water body of the similar small sample data according to the initial conditions; Calculate the rate of decrease of antibiotic concentration in water between two adjacent time points; The least squares method was used to perform a segmented linear regression analysis on the rate of decrease of the antibiotic concentration in the water body, and the stage transition points with significantly different slopes were identified through the F test and used as critical points; According to the critical points under different conditions, all data in the similar small sample data from the initial moment to the critical point are divided into first-stage data, and the data after that are divided into second-stage data.

5. The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon according to claim 1, characterized in that: The method of generating extended data based on the second-stage data transfer learning and screening high-confidence samples from the extended data in combination with the rapid adsorption range includes: The temporal convolutional network (TCN) is used to extract the dependent features of the water antibiotic concentration, water pH value, water temperature and plant height in the historical data to generate a temporal feature matrix. , where m is the initial number of water bodies, t is the length of the time series, and d is the data dimension. The time series feature matrix is ​​mean pooled to convert each time series feature into a statistical feature vector of fixed dimension. ; The singular value decomposition method is used to reduce the dimension of the statistical characteristic matrix formed by the statistical characteristic vectors, and the first k principal components are retained as subspace basis vectors based on the cumulative variance contribution rate greater than 95%. ; Calculate the statistical eigenvector With the subspace basis vector The feature importance weight of is calculated as follows: ,in is the inner product operation; The extended feature vector is generated based on the statistical feature vector, subspace basis vector and feature importance weight. The calculation formula is as follows: ； in, is the extended eigenvector, is the inner product operation, The target feature vector to be expanded is composed of the data to be predicted. are the subspace basis vectors, is the feature importance weight, is the statistical eigenvector, is a random perturbation vector that satisfies physical constraints; The maximum mean difference is used to evaluate the distribution difference between the extended feature vector and the original historical data. If the maximum mean difference is less than the preset difference threshold, the extended feature vector is retained; Repeatedly generating multiple extended feature vectors to form extended data; The initial antibiotic concentration of the extended data is extracted, and the antibiotic removal rate of the water body under different initial conditions of the extended data is calculated based on the initial antibiotic concentration. The samples with removal rates outside the rapid adsorption range are eliminated, and the antibiotic concentration of the initial data is tested according to the 3σ principle to screen out high-confidence samples.

6. The method for predicting the removal effect of antibiotics from water based on the synergy of plants and activated carbon according to claim 1, characterized in that: The method of constructing a charcoal plant adsorption model based on the high-confidence sample and using the long-ying oat optimization algorithm (AOO) to perform hyperparameter optimization includes: A GRU neural network model was constructed, using the root mean square error as the fitness function. The input layer dimension was the sum of the dimension of the data to be predicted and the dimension of the time parameter. It contained two layers of GRU units, the activation function was a linear function, and the output was the predicted value of the antibiotic concentration in the water body. AOO initialization uses Tent chaos mapping to initialize the population, and each individual is a combination of the number of hidden layer neurons, learning rate, Dropout ratio and weight matrix initialization; Generate a random number between 0 and 1 ,when The update formula of oat position in the exploration phase is: ； in and The tth and The position of the i-th oat individual in generation, is the population size, is the dynamic adjustment factor, is the maximum number of iterations, is a random vector of 0 to 1 with the same dimension dim among the individuals in the population, It is Hadamard. is the upper bound vector of the target problem, is the position of the best individual in the group; otherwise The update formula of oat position in the development stage is: ； in Indicates the quality of oat seeds, Represents the operation of independently generating random numbers from 0 to 1, is a randomly generated value between 0 and 1, dim represents the dimension of the target problem, Eccentricity coefficient during seed rolling, Indicates the length of the main awn of oat seeds, express, Indicates generating a random matrix with the same dimension as dim and values ​​between -A and A. , , is the result of the Lévy flight function with parameter N,dim,1.5; The algorithm is terminated when the maximum number of iterations is reached, and the optimal hyperparameter combination corresponding to the global optimal position is output; The optimal hyperparameter combination is used to input a GRU neural network model, the extended data set is used for training, and the trained GRU neural network model is used as a carbon plant adsorption model.

Images