A method for predicting harmful algal blooms based on Tent-GWO-GRU

Abstract

This application discloses a method for predicting harmful algal blooms based on Tent-GWO-GRU, comprising: acquiring and preprocessing historical water quality data of the target watershed; constructing a gated recurrent unit neural network (GRU) model based on the historical water quality data; optimizing the Grey Wolf Algorithm (GWO) based on the Tent chaotic mapping algorithm; and optimizing the GRU model parameters and making predictions. This application applies a time delay to the index data, fully considering the lag in algal bloom growth and evolution, thus improving the feasibility of the method. The introduction of the Tent chaotic mapping algorithm improves the method for generating the initial wolf pack in the GWO algorithm, resulting in a more uniform distribution of the initial wolf pack, significantly improving the algorithm's fitness value, and making it easier to find the global optimum. This application uses the Tent-GWO optimization algorithm to optimize the hyperparameters and network structure of the GRU model, which not only improves the model's stability but also achieves higher prediction accuracy.

Description

Technical Field

[0001] This application belongs to the field of harmful algal bloom technology, specifically involving a method for predicting harmful algal blooms based on Tent-GWO-GRU. Background Technology

[0002] Harmful algal blooms (HABs) are ecological disasters caused by the rapid proliferation or massive accumulation of planktonic algae in water. In recent decades, due to eutrophication, global warming, maritime transport, and the development of mariculture, the frequency, duration, scope, and severity of harmful algal blooms have been steadily increasing globally. To mitigate the potential losses from harmful algal blooms, it is essential to establish effective models to predict the growth and evolution of algae.

[0003] In recent years, machine learning models have been widely used in water quality prediction modeling. Models such as Artificial Neural Networks (ANN), Recurrent Neural Networks (GRU), Extreme Learning Machines (ELM), Random Forests (RF), and Support Vector Machines (SVM) have been extensively used for regression prediction. Machine learning models bypass the complex mechanisms of algal bloom growth and evolution, focusing on the mapping relationship between system output and input. By learning from large amounts of historical data, they can accurately approximate highly nonlinear relationships. Mamun et al. used SVM to predict algal chlorophyll a and water transparency in a clean water ecosystem. Experimental results showed that the SVM model had poor prediction accuracy and stability, and could not accurately predict the growth and evolution of harmful algal blooms. Lu Yuanbo et al. used a Recurrent Neural Network (GRU) model to predict water quality at cross-border sections in Guangdong Province with high accuracy. To facilitate the transfer and use of the GRU model, they proposed the PSO-GRU water quality prediction model, introducing a particle swarm optimization algorithm to optimize the hyperparameters of the GRU, reducing the empirical and random nature of hyperparameter settings and improving the model's prediction accuracy. However, the convergence speed of the Particle Swarm Optimization (PSO) algorithm is relatively slow. While the Gray Wolf Optimization (GRU) algorithm can address this issue better, it suffers from the problem of relatively random initial point selection. Therefore, Li Jin et al. utilized the randomness and ergodicity of chaotic systems to generate a more uniformly distributed Tent chaotic sequence to initialize the gray wolf population, obtaining a more uniformly distributed and diverse initial solution, thus improving the algorithm's global search capability. However, how to apply Tent-GWO and GRU to the field of harmful algal blooms still requires further exploration. Summary of the Invention

[0004] This application provides a method for predicting harmful algal blooms based on Tent-GWO-GRU to solve the above-mentioned technical problems.

[0005] To solve the above-mentioned technical problems, this application adopts the following technical solution: a method for predicting harmful algal blooms based on Tent-GWO-GRU, comprising the following steps:

[0006] Acquire historical water quality data for the target watershed and perform preprocessing;

[0007] Based on historical water quality data, a gated recurrent unit neural network GRU model was constructed.

[0008] Based on the Tent chaotic mapping algorithm, the Grey Wolf Algorithm (GWO) is optimized.

[0009] Based on the Gated Recurrent Unit (GRU) neural network model and the Grey Wolf (GWO) algorithm, the parameters of the GRU model are optimized and predictions are made.

[0010] Furthermore, methods for acquiring and preprocessing historical water quality data of the target watershed include:

[0011] Historical water quality data of the target watershed were collected, and chlorophyll concentration was selected as an indicator to measure the degree of algal growth.

[0012] Based on historical water quality data and factors affecting algal growth, the first feature dataset was obtained;

[0013] Based on linear interpolation, missing values ​​and outliers are detected in the water quality feature dataset to obtain the second feature dataset.

[0014] Perform range standardization on the indicator data in the second feature dataset to obtain the third feature dataset;

[0015] Preprocessed data is obtained based on time delay and indicator data in the third feature dataset;

[0016] The preprocessed data is divided into training and testing sets according to a preset ratio.

[0017] Furthermore, the method for constructing a gated recurrent unit (GRU) neural network includes:

[0018] The indicator data in the preprocessed data is reorganized to obtain a vector set;

[0019] Based on the vector set, a reset gate is constructed for the gated recurrent unit (GRU) neural network model.

[0020] Based on the reset gate of the GRU (Gated Recurrent Unit) neural network model, obtain the update gate of the GRU model.

[0021] Based on the update gate of the GRU (Gated Recurrent Unit) neural network model, obtain the fully connected layer of the GRU model.

[0022] The output value of the GRU neural network is obtained by using a fully connected layer based on the GRU model.

[0023] Furthermore, based on the Tent chaotic mapping algorithm, the method for optimizing the Grey Wolf Algorithm (GWO) includes:

[0024] Based on the Tent chaotic mapping algorithm, the wolf pack in the Grey Wolf Algorithm (GWO) is initialized to obtain the initial wolf pack parameters and the initial number of iterations.

[0025] Based on the initial wolf pack parameters, obtain the fitness value of each individual gray wolf;

[0026] Compare the fitness values ​​of all individual gray wolves to obtain the first, second, and third best solutions;

[0027] Based on the optimal solution gray wolf individuals corresponding to the first, second, and third optimal solutions, obtain the distance parameters between the remaining gray wolf individuals and the optimal solution gray wolf individuals;

[0028] Update the parameter vector based on the distance parameter;

[0029] Repeat the above steps until the number of iterations reaches the preset value to obtain the optimal gray wolf parameters.

[0030] Furthermore, methods for optimizing GRU model parameters and making predictions include:

[0031] The data in the training set is input into the gated recurrent unit neural network (GRU) model to obtain the predicted value of chlorophyll concentration and the prediction accuracy.

[0032] Based on the true and predicted values ​​of chlorophyll concentration, obtain the model loss function;

[0033] The model loss function is optimized based on the Adam optimization algorithm to obtain a pre-trained model.

[0034] Using the loss function as the fitness function, the learning rate, number of hidden layer nodes, and regularization coefficient in the pre-trained model are optimized according to the optimized Tent-GWO algorithm to obtain the pre-trained Tent-GWO-GRU model.

[0035] The chlorophyll concentration in the test set was obtained based on the Tent-GWO-GRU model.

[0036] The beneficial effects of this application are as follows: This application applies a time delay to the index data, fully considering the delayed nature of algal bloom growth and evolution, thus improving the feasibility of the method. The introduction of the Tent chaotic mapping algorithm improves the method of generating the initial wolf pack in the GWO algorithm, resulting in a more uniform distribution of the initial wolf pack, significantly improving the algorithm's fitness value, and making it easier to find the global optimum. Furthermore, this application employs the Tent-GWO optimization algorithm to optimize the hyperparameters and network structure of the GRU model, which not only improves the model's stability but also achieves higher prediction accuracy. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating an embodiment of the harmful algal bloom prediction method based on Tent-GWO-GRU of this application;

[0038] Figure 2 for Figure 1 A flowchart illustrating step S1 of an embodiment;

[0039] Figure 3 for Figure 1 A flowchart illustrating step S2 of an embodiment;

[0040] Figure 4 yes Figure 3 A schematic diagram of the unit structure of an embodiment of the GRU (Gated Recurrent Unit) neural network model in China;

[0041] Figure 5 for Figure 1 A flowchart illustrating step S3 of an embodiment;

[0042] Figure 6 for Figure 1 A flowchart illustrating step S4 in one embodiment;

[0043] Figure 7 This is a comparison chart of the fitness of the improved GWO and the original GWO in an embodiment of the harmful algal bloom prediction method based on Tent-GWO-GRU in this application.

[0044] Figure 8 This is a comparison chart of the goodness of fit between the harmful algal bloom prediction method and the SVM model in one embodiment of the harmful algal bloom prediction method based on Tent-GWO-GRU in this application.

[0045] Figure 9 This is a model prediction effect diagram of a harmful algal bloom prediction method based on Tent-GWO-GRU in this application.

[0046] Figure 10 This is a diagram showing the model's performance in predicting harmful algal blooms based on the SVM model. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0048] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0049] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the harmful algal bloom prediction method based on Tent-GWO-GRU of this application. The method includes:

[0050] Step S1. Obtain historical water quality data for the target watershed and perform preprocessing.

[0051] For details, please refer to Figure 2 , Figure 2 for Figure 1 A flowchart illustrating an embodiment of step S1. Step S1 includes:

[0052] Step S11. Collect historical water quality data of the target watershed and select chlorophyll concentration as an indicator to measure the degree of algal growth.

[0053] Specifically, water quality data from January 1, 2000 to December 31, 2023 were collected from the official websites of monitoring stations in the target watershed. Chlorophyll-a (Chl-a) concentration was selected as an indicator to measure the degree of algal growth. From the factors that directly affect algal growth, eight indicators were selected as model inputs: total inorganic nitrogen (TIN) concentration, total phosphorus (PO4) concentration, chlorophyll-a (Chl-a) concentration, dissolved oxygen (DO) concentration, water temperature (Temp), Sage disc depth (SSD), 5-day biological oxygen demand (BOD5), and pH value.

[0054] Step S12. Based on historical water quality data and factors affecting algae growth, obtain the first feature dataset.

[0055] Specifically, water quality data from January 1, 2000 to December 31, 2023 were integrated based on selected indicators to obtain the first feature dataset. .

[0056] Step S13. Based on the linear interpolation method, perform missing value and outlier checks on the water quality feature dataset to obtain the second feature dataset.

[0057] Specifically, for the first feature dataset Missing and outlier detection was performed, and linear interpolation was used for imputation to obtain the second feature dataset. .

[0058] Step S14. Perform range standardization on the indicator data in the second feature dataset to obtain the third feature dataset.

[0059] Specifically, for the second feature dataset The eight indicator data points were standardized using the range, as shown in the following formula:

[0060] ;

[0061] in, and These represent the original value and normalized value of the variable, respectively. and These represent the maximum and minimum values ​​of the variable, respectively.

[0062] Let the standardized third feature dataset be... for:

[0063] ;

[0064] in, These represent the normalized values ​​of the eight indicator data.

[0065] Step S15. Based on the time delay and the indicator data in the third feature dataset, obtain the preprocessed data.

[0066] Specifically, to accurately predict the growth and evolution of harmful algal blooms, the third feature dataset... The eight indicators in the text introduce a time lag of 7 to 13 days respectively. , 8, 9, 10, 11, 12, 13) Construct preprocessed data containing 56 input variables; wherein, the preprocessed data is:

[0067] ;

[0068] in, The period is from January 14, 2000 to December 31, 2023.

[0069] Step S16. Divide the preprocessed data into training set and test set according to a preset ratio.

[0070] Specifically, the preprocessed data is divided into training and test sets in a 7:3 ratio.

[0071] Step S2. Construct a gated recurrent unit neural network (GRU) model based on historical water quality data.

[0072] For details, please refer to Figure 3 , Figure 3 for Figure 1 A flowchart illustrating step S2 of an embodiment. Step S2 includes:

[0073] Step S21. Reorganize the indicator data in the preprocessed data to obtain a vector set.

[0074] Specifically, the dataset The 56 indicator data points were reorganized to obtain a vector set;

[0075] ;

[0076] in, It is a vector set.

[0077] Step S22. Based on the vector set, construct the reset gate of the gated recurrent unit neural network (GRU) model.

[0078] Specifically, based on formula (1), a reset gate is constructed for the gated recurrent unit (GRU) neural network model; where formula (1) is:

[0079] (1);

[0080] Among them, the reset door The smaller the value, the more it indicates The more information you need to forget at any time; and Both are weight matrices. To reset the gate's bias vector, express The output of the hidden layer at any given time.

[0081] Step S23. Based on the reset gate of the GRU (Gated Recurrent Unit) neural network model, obtain the update gate of the GRU model.

[0082] For details, please refer to Figure 4 Based on formulas (2) to (4), the update gate of the gated recurrent unit neural network (GRU) model is constructed; where formulas (2) to (4) are:

[0083] (2);

[0084] (3);

[0085] (4);

[0086] Among them, the reset door The smaller the value, the more historical information is retained. This represents the candidate hidden layer state. express The state vector output at each time step. and Both are weight matrices. and For bias vectors, This indicates that a dot product is performed at the corresponding position.

[0087] Step S24. Based on the update gate of the GRU (Gated Recurrent Unit) neural network model, obtain the fully connected layer of the GRU model.

[0088] Specifically, based on formula (5), a fully connected layer of the gated recurrent unit neural network (GRU) model is constructed; where formula (5) is:

[0089] (5);

[0090] in, For the fully connected layer The output of each neuron; Indicates the weights of the fully connected layer; represents the bias value of the fully connected layer; f(x) represents the ReLU activation function.

[0091] Step S25. Obtain the output value of the GRU neural network based on the fully connected layer of the GRU model.

[0092] Specifically, the output layer of a gated recurrent unit (GRU) neural network is a real-valued regression of a single neuron. The output value can be expressed as formula (6):

[0093] (6);

[0094] in, Represents the normalized weights. This represents the bias value.

[0095] Step S3. Optimize the Grey Wolf Algorithm (GWO) based on the Tent chaotic mapping algorithm.

[0096] For details, please refer to Figure 5 , Figure 5 for Figure 1 A flowchart illustrating step S3 of an embodiment. Step S3 includes:

[0097] Step S31. Based on the Tent chaotic mapping algorithm, initialize the wolf pack in the Grey Wolf Algorithm (GWO) and obtain the initial wolf pack parameters and the initial number of iterations.

[0098] Specifically, the Tent chaotic mapping algorithm is used to initialize the wolf pack, generating... The specific formula for determining the location of an individual gray wolf is as follows:

[0099] ;

[0100] in, This indicates the original location of the individual gray wolf. Indicates the position after the Tent chaotic mapping. The wolf pack parameters a, A, and C are initialized as follows:

[0101] ;

[0102] ;

[0103] Where C is the oscillation factor and A is the convergence factor. and The number 'a' is a random number between [0,1], and 'a' decreases linearly from 2 to 0 as the number of iterations increases. The maximum number of iterations, This indicates the k-th iteration.

[0104] Step S32. Based on the initialized wolf pack parameters, obtain the fitness value of each individual gray wolf.

[0105] Step S33. Compare the fitness values ​​of all individual gray wolves to obtain the first, second, and third optimal solutions.

[0106] Specifically, the fitness values ​​of individuals are compared to determine the current best solution. Second optimal solution and the third optimal solution .

[0107] Step S34. Based on the optimal solution gray wolf individuals corresponding to the first optimal solution, the second optimal solution, and the third optimal solution, obtain the distance parameters between the remaining gray wolf individuals and the optimal solution gray wolf individuals.

[0108] Specifically, based on the swing factor C, the comparison between other gray wolf individuals in the population and the optimal gray wolf individual is calculated. , Distance parameters , , .

[0109] Step S35. Update the parameter vector based on the distance parameter.

[0110] Specifically, based on the distance parameter , , And the convergence factor A is used to update the position of each gray wolf, and to update the wolf pack parameters a, A, and C.

[0111] Step S36. Repeat the above steps until the number of iterations reaches the preset value to obtain the optimal gray wolf parameters.

[0112] Specifically, determine the current iteration number. Has the maximum number of iterations been reached? ,like Return to step S32; otherwise, determine the optimal gray wolf parameters.

[0113] Step S4. Based on the Gated Recurrent Unit (GRU) neural network model and the Grey Wolf Algorithm (GWO), optimize the GRU model parameters and make predictions.

[0114] For details, please refer to Figure 6 , Figure 6 for Figure 1 A flowchart illustrating step S4 of an embodiment. Step S4 includes:

[0115] Step S41. Input the data from the training set into the gated recurrent unit neural network (GRU) model to obtain the predicted value of chlorophyll concentration and the prediction accuracy.

[0116] Specifically, the training set data and the corresponding chlorophyll concentration at the time are input into the gated recurrent unit neural network (GRU) model, which outputs the predicted chlorophyll concentration and the prediction accuracy.

[0117] Step S42. Obtain the model loss function based on the true and predicted values ​​of chlorophyll concentration.

[0118] Specifically, based on the predicted and actual values ​​of chlorophyll concentration, the model loss function RMSE of the GRU model is calculated; the model loss function RMS is shown below:

[0119] ;

[0120] in, Indicates the first Predicted values ​​of chlorophyll concentration for each sample. Indicates the first The true value of chlorophyll concentration for each sample This represents the total number of samples.

[0121] Step S43. Optimize the model loss function based on the Adam optimization algorithm to obtain the pre-trained model.

[0122] Specifically, the Adam optimization algorithm is used to optimize the model loss function to obtain the pre-trained model.

[0123] Step S44. Using the loss function as the fitness function, optimize the learning rate, number of hidden layer nodes, and regularization coefficient in the pre-trained model according to the optimized Tent-GWO algorithm to obtain the pre-trained Tent-GWO-GRU model.

[0124] Step S45. Based on the Tent-GWO-GRU model, obtain the chlorophyll concentration in the test set.

[0125] See Figure 7-10 To verify the effectiveness of this application, numerical simulation using MATLAB was conducted. This example obtained Hong Kong water quality data from January 1, 1988 to December 31, 2018 from the weakest flushing monitoring station of the Hong Kong Conservation Department. Linear interpolation was used for data preprocessing, and time delays of 7 to 13 days were introduced for each of the eight indicators. Based on the above data, in the Tent-GWO-GRU model proposed in this invention, the gray wolf population size is 5, and the number of iterations is 20. Figure 7 These are the fitness curves for the improved GWO algorithm and the original GWO algorithm. The optimal fitness of the Tent-GWO algorithm is 0.0432, while the optimal fitness of the original GWO algorithm is 0.0475. Figure 7 It can be observed that the improved GWO algorithm has better convergence speed and accuracy than the original GWO algorithm, indicating that the improved GWO algorithm has better global search capability than the original GWO algorithm. A comparison of the goodness-of-fit of the improved GWO-GRU model and the SVM model on the test set is shown below. Figure 8 As shown in the figure, the Tent-GWO-GRU model can better capture the growth trend and change magnitude of ChI-a concentration in the test set, with fewer outliers. Figure 9 , Figure 10 The figures show the fitting results of the Tent-GWO-GRU and SVM predictions to the actual values, respectively. It can be seen from the figures that the Tent-GWO-GRU model significantly outperforms the SVM model in terms of fitting performance. The Tent-GWO-GRU model proposed in this invention has good predictive performance.

[0126] This application incorporates a time lag in the index data, fully considering the delayed nature of algal bloom growth and evolution, thus improving the feasibility of the method. The Tent chaotic mapping algorithm is introduced to improve the method of generating the initial wolf pack in the GWO algorithm, resulting in a more uniform distribution of the initial wolf pack, significantly improving the algorithm's fitness value and making it easier to find the global optimum. Furthermore, this application employs the Tent-GWO optimization algorithm to optimize the hyperparameters and network structure of the GRU model, which not only improves the model's stability but also achieves higher prediction accuracy.

[0127] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for predicting harmful algal blooms based on Tent-GWO-GRU, characterized in that, Includes the following steps: Historical water quality data of the target watershed is acquired and preprocessed. Historical water quality data of the target watershed is collected, and chlorophyll concentration is selected as an indicator to measure the degree of algal growth. Based on the historical water quality data and factors affecting algal growth, a first feature dataset is obtained; Based on linear interpolation, missing values ​​and outliers are detected in the water quality feature dataset to obtain the second feature dataset. The index data in the second feature dataset are subjected to range standardization to obtain the third feature dataset; Based on the time delay and the indicator data in the third feature dataset, preprocessed data is obtained; The preprocessed data is divided into a training set and a test set according to a preset ratio; Based on the historical water quality data, a gated recurrent unit neural network (GRU) model is constructed. Based on the Tent chaotic mapping algorithm, the Grey Wolf Algorithm (GWO) is optimized. Based on the Gated Recurrent Unit (GRU) neural network model and the Grey Wolf Algorithm (GWO), the GRU model parameters are optimized and predictions are made: The data in the training set is input into the gated recurrent unit neural network (GRU) model to obtain the predicted value of chlorophyll concentration and the prediction accuracy. Based on the actual and predicted values ​​of chlorophyll concentration, the model loss function is obtained; The model loss function is optimized using the Adam optimization algorithm to obtain a pre-trained model. Using the loss function as the fitness function, the learning rate, number of hidden layer nodes, and regularization coefficient in the pre-trained model are optimized according to the optimized Tent-GWO algorithm to obtain the pre-trained Tent-GWO-GRU model. Based on the Tent-GWO-GRU model, the chlorophyll concentration in the test set is obtained.

2. The method according to claim 1, characterized in that, The method for constructing a gated recurrent unit neural network (GRU) includes: The indicator data in the preprocessed data are reorganized to obtain a vector set; Based on the vector set, a reset gate is constructed for the Gated Recurrent Unit (GRU) neural network model. Based on the reset gate of the GRU (Gated Recurrent Unit) neural network model, the update gate of the GRU model is obtained. Based on the update gate of the GRU (Gated Recurrent Unit) neural network model, the fully connected layer of the GRU model is obtained. The output value of the GRU neural network is obtained based on the fully connected layer of the GRU model.

3. The method according to claim 2, characterized in that, The method for optimizing the Grey Wolf (GWO) algorithm based on the Tent chaotic mapping algorithm includes: Based on the Tent chaotic mapping algorithm, the wolf pack in the Grey Wolf Algorithm (GWO) is initialized to obtain the initial wolf pack parameters and the initial number of iterations. Based on the initial wolf pack parameters, the fitness value of each individual gray wolf is obtained; Compare the fitness values ​​of all the gray wolf individuals to obtain the first, second, and third optimal solutions; Based on the first optimal solution, the second optimal solution, and the optimal solution gray wolf individuals corresponding to the third optimal solution, obtain the distance parameters between the remaining gray wolf individuals and the optimal solution gray wolf individuals; Update the parameter vector based on the distance parameter; Repeat the above steps until the number of iterations reaches a preset value to obtain the optimal gray wolf parameters.

Images