Water environment index change trend prediction and model establishment method, equipment and medium

Abstract

The invention provides a water environment index change trend prediction and model establishment method and device, and a medium, and the model establishment method comprises the following steps: obtaining historical index data of a target water environment and driving influence factor data of the historical index data, and enabling the historical index data and the driving influence factor data to be used for establishing a training data set; hyper-parameters of the water environment index change trend prediction model are determined by adopting an automatic optimization algorithm, and the key hyper-parameters comprise a learning rate, the size of a convolution kernel and the number of neurons; the training data set is used for training a water environment index change trend prediction model, and a basic network model of the water environment index change trend prediction model is a combination of a convolutional neural network, a long and short term memory neural network and a self-attention mechanism. The method is based on the CNN-LSTM-Attention neural network model and adopts the automatic optimization algorithm to adjust the hyper-parameters, so that the parameter setting optimality in the model training process is ensured, higher adaptability and stability are obtained, and a more reliable water environment index change trend prediction result can be provided.

Description

Technical Field

[0001] The present invention relates to the field of water environment indicator change trend prediction, and specifically, to a water environment indicator change trend prediction model establishment method, a water environment indicator change trend prediction method, a water environment indicator change trend prediction model, and equipment and computer-readable storage medium for implementing the water environment indicator change trend prediction model establishment method and the water environment indicator change trend prediction method. Background Art

[0002] In today's era of information explosion, data prediction technology plays an increasingly important role in various fields. Especially in industries such as finance, meteorology, and energy, accurate predictions can help decision makers develop more scientific strategies and plans. As an important branch of data prediction, time series prediction aims to predict future trends or behavior patterns by analyzing historical data. Existing time series prediction methods mainly include statistical models (such as ARIMA), machine learning methods (such as support vector machines SVM), and deep learning techniques. Although these methods can provide effective predictions in some cases, they still have some limitations. For example, traditional statistical models usually assume a specific form of data distribution, which limits their ability to handle complex data sets. Although machine learning methods perform well on some problems, they usually require manual feature engineering, which is not only time-consuming and labor-intensive, but may also miss important information. Deep learning methods, especially recurrent neural networks (RNNs) and their variants, such as long short-term memory networks (LSTMs), have made significant progress in processing sequence data, but these models often require a lot of parameter adjustment and are prone to falling into local optimality.

[0003] In addition, as the complexity of the model increases, it becomes increasingly difficult to select the appropriate network structure and hyperparameters. The setting of hyperparameters has a significant impact on model performance, but finding the optimal hyperparameter combination often requires a lot of experiments and time. Traditional grid search or random search methods are inefficient in high-dimensional parameter spaces and it is difficult to guarantee the global optimal solution. Summary of the invention

[0004] The purpose of the present invention is to solve at least one of the above-mentioned deficiencies in the prior art. For example, one of the purposes of the present invention is to improve a method for establishing a water environment indicator change trend prediction model based on automatic optimization algorithm to optimize hyperparameters and deep learning network.

[0005] In order to achieve the above-mentioned objectives, the present invention provides, on the one hand, a method for establishing a water environment indicator change trend prediction model.

[0006] The method for establishing a water environment indicator change trend prediction model comprises the following steps:

[0007] S1. Obtain historical indicator data of the target water environment and driving influencing factor data of the historical indicator data for establishing a training data set, wherein the driving influencing factor data include meteorological data, hydrological data, domestic pollution emission data, industrial pollution emission data, agricultural pollution emission data, endogenous pollution emission data and water body self-purification capacity data.

[0008] S2. Use an automatic optimization algorithm to confirm the hyperparameters of the water environment indicator change trend prediction model, wherein the key hyperparameters include learning rate, convolution kernel size and number of neurons.

[0009] S3. Using the training data set to train the water environment index change trend prediction model, wherein the basic network model of the water environment index change trend prediction model is a combination of a convolutional neural network, a long short-term memory neural network and a self-attention mechanism.

[0010] In an exemplary embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention, the network structure of the water environment indicator change trend prediction model can be a sequence input layer, a convolution layer, a ReLU activation layer, a sequence anti-folding layer, an LSTM layer, a self-attention mechanism layer, a fully connected layer and a regression layer connected in sequence.

[0011] In an exemplary embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention, the automatic optimization algorithm may be a swarm intelligence optimization algorithm that simulates automatic hunting behavior.

[0012] In an exemplary embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention, in step S3, the water environment indicator change trend prediction model can be trained using an Adam optimizer or an SGD optimization algorithm and corresponding first training parameters.

[0013] In an exemplary embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention, in step S3, the weight of the water environment indicator change trend prediction model in training can be optimized by a back propagation algorithm.

[0014] In an exemplary embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention, in step S3, the error index, determination coefficient or correlation coefficient of the water environment indicator change trend prediction model can be calculated through a pre-established test set to evaluate the performance of the water environment indicator change trend prediction model.

[0015] Another aspect of the present invention provides a water environment indicator change trend prediction model, which is obtained according to any of the above-mentioned water environment indicator change trend prediction model establishment methods.

[0016] Another aspect of the present invention provides a method for predicting the changing trend of water environment indicators, which may include: using the above-mentioned water environment indicator changing trend prediction model to predict the changing trend of water environment indicators of the target water environment.

[0017] In another aspect, the present invention provides a computer device, the computer device comprising:

[0018] A processor; a memory storing a computer program, which, when executed by the processor, implements at least one of the water environment indicator change trend prediction model establishment method and the water environment indicator change trend prediction method.

[0019] On the other hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements at least one of the water environment indicator change trend prediction model establishment method and the water environment indicator change trend prediction method as described above.

[0020] Compared with the prior art, the beneficial effects of the present invention include at least one of the following:

[0021] (1) The present invention automatically adjusts the hyperparameters of the deep learning network, such as the learning rate, convolution kernel size, and number of neurons, through an automatic optimization algorithm to ensure the optimality of parameter settings during model training. This automated hyperparameter adjustment not only saves a lot of time and effort for manual adjustment, but also improves the accuracy and efficiency of model training. Compared with traditional experience-based or grid search methods, the automatic optimization algorithm can converge to a better solution more quickly and significantly improve the performance of the model.

[0022] (2) The present invention introduces a self-attention mechanism layer, which enables the deep learning network to capture long-distance dependencies in time series data, thereby significantly improving the model's generalization ability to new data. This mechanism is particularly suitable for predicting the trend of changes in water environment indicators because it can better process and predict water environment indicator trend data with complex patterns and seasonal changes. Compared with models that rely only on short-term historical data, the method of the present invention can more accurately predict changes in future water environment indicator trends.

[0023] (3) The deep learning network structure of the present invention is carefully designed, including convolutional layers, LSTM layers and self-attention layers. The combination of these layers enables the model to extract richer features from the training data, perform effective feature fusion, and make full use of the advantages of different network structures to process complex features and long-term dependencies in time series data. The optimal hyperparameters obtained by the automatic optimization algorithm further enhance the learning ability of the model, making the prediction results closer to the true value, thereby showing higher accuracy than the prior art in multiple error evaluation indicators.

[0024] (4) While ensuring the prediction accuracy, the present invention effectively controls the computational complexity of the model through reasonable hyperparameter settings and network structure design. This not only reduces the computing resources required in the training and prediction process, such as memory and processor time, but also speeds up the response speed of the model, making the method of the present invention more suitable for real-time or near real-time water environment indicator change trend prediction application scenarios, which is of great significance for the operation and management of water environment systems that require rapid decision-making.

[0025] (5) The iterative search process of the automatic optimization algorithm of the present invention can find robust hyperparameter combinations under a variety of different data distributions and noise conditions. This robustness enables the model to not only perform well on training data, but also adapt to various uncertainties and changes in actual operation, such as weather changes, holiday effects, etc. Compared with the model with fixed hyperparameter settings, the method of the present invention has stronger adaptability and stability, and can provide more reliable prediction results of water environment indicator change trends. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and other objects and / or features of the present invention will become more apparent through the following description in conjunction with the accompanying drawings, in which:

[0027] Figure 1 A schematic flow chart showing the steps of an exemplary embodiment of a method for establishing a water environment indicator change trend prediction model of the present invention.

[0028] Figure 2 A schematic diagram showing the results of predicting water environment indicator change trends according to an embodiment of the method for establishing a water environment indicator change trend prediction model of the present invention is shown. DETAILED DESCRIPTION

[0029] Hereinafter, the method, device and medium for predicting the change trend of water environment indicators and establishing a model according to the present invention will be described in detail in conjunction with exemplary embodiments.

[0030] It should be noted that the terms "S1", "S2", "S3"... and "first", "second", "third"... etc. used in the present invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0031] In the existing technology, most of the predictions are made for water quality parameters, such as ammonia nitrogen index, potassium permanganate index, total nitrogen index and total phosphorus index. However, water quality indicators can only be directly used to evaluate water quality conditions, identify pollution sources, and provide pollution warnings. However, in order to achieve sustainable development of the water environment, it is necessary to formulate basin restoration plans, load reduction targets and pollution control measures. Therefore, a model is needed to predict the changing trends of water environment indicators.

[0032] In order to solve the above problems, the present invention proposes a method for establishing a water environment indicator change trend prediction model. The model uses a deep network to more effectively capture the nonlinear characteristics, local patterns and long-term dependencies in time series data, so as to improve the prediction accuracy and generalization ability of the model. In addition, the hyperparameters are intelligently searched through an automatic optimization algorithm, which avoids the tedious process of traditional manual parameter adjustment and reduces the number of iterations required to find the optimal solution, thereby accelerating the convergence speed of the model and improving the efficiency of model training.

[0033] Exemplary Embodiment 1

[0034] In this exemplary embodiment, Figure 1 As shown, the method for establishing a water environment index change trend prediction model includes the following steps:

[0035] S1. Obtain historical indicator data of the target water environment and driving and influencing factor data of the historical indicator data for establishing a training data set, wherein the driving and influencing factor data of the historical indicator data include meteorological data, hydrological data, domestic pollution emission data, industrial pollution emission data, agricultural pollution emission data, endogenous pollution emission data and water body self-purification capacity data.

[0036] Optionally, during the data preprocessing stage, the acquired historical indicator data and the driving influencing factor data of the historical indicator data can be preliminarily analyzed, including data cleaning and removal of outliers. It is also possible to determine the required prediction time step, data type and time span by performing time series analysis on the historical indicator data and the driving influencing factor data of the historical indicator data, and import the preprocessed historical indicator data and its driving influencing factor data into the training data set. Among them, determining the required prediction time step can be used to confirm the trend of all indicator changes in the next 1 to n days, determining the data type can be used to eliminate parameters that have no effect on the prediction results but consume a lot of model training time, and determining the time span can be used to avoid using too much data for training and wasting storage and computing resources.

[0037] Optionally, you can clear the memory and screen during the data preprocessing phase to ensure a clean computing environment.

[0038] Optionally, different data cleaning techniques, such as moving average filtering and wavelet transform, may be used to adapt to different data noise characteristics.

[0039] Optionally, in the data preprocessing stage, the acquired historical indicator data and its driving influencing factor data are normalized.

[0040] Furthermore, different normalization methods can be selected to adapt to the model's specific requirements for data distribution, such as Z-score normalization or maximum and minimum normalization methods.

[0041] Furthermore, the historical indicator data, the driving factors of the historical indicator data, and the actual output data of the water environment indicator change trend prediction model can also be normalized, with the [0,1] interval as the target, so that the data meets the input requirements of the deep learning model (i.e., the water environment indicator change trend prediction model). For example, the "mapminmax" function can be used to normalize the data to improve the efficiency and accuracy of model training.

[0042] S2. Use the automatic optimization algorithm to confirm the hyperparameters of the water environment indicator change trend prediction model, among which the key hyperparameters include learning rate, convolution kernel size and number of neurons.

[0043] The selection of the initial learning rate usually needs to vary according to the optimization algorithm used to train the water environment indicator change trend prediction model. For example, if the SGD optimization algorithm is selected, the initial learning rate can be [1e-2, 1e-1], while for the Adam optimization algorithm, the initial learning rate can be [1e-3, 1e-2].

[0044] 1D Convolutional Layer is often used to process sequence data, such as time series or text data. For the convolution kernel size, a small convolution kernel can be used, and its size such as 1x1, 3x1 or 5x1 is a common choice. 1x1 convolution kernel is often used to reduce dimension or add nonlinearity in deep networks without adding too much computational burden. 3x1 or 5x1 convolution kernels can capture local sequence patterns while maintaining computational efficiency. Medium convolution kernels can also be used. In some cases, 7x1 or larger convolution kernels may be used to capture dependencies of longer sequences.

[0045] As for the number of neurons, for example, 8, 16, 32, 64, 128...

[0046] Optionally, the parameters of the automatic optimization algorithm can be adjusted, such as the number of search individuals, the maximum number of iterations, the lower and upper bounds of the variables, to obtain better optimization results. It is also possible to find the hyperparameter combination that minimizes the prediction error of the indicator change trend through iterative search.

[0047] Optionally, the optimized hyperparameters can be used to set the training options of the water environment indicator trend prediction model, such as learning rate, batch size, and learning rate reduction strategy.

[0048] Optionally, the network structure of the water environment index change trend prediction model can be a sequence input layer, a convolution layer, a ReLU activation layer, a sequence unfolding layer, an LSTM layer, a self-attention mechanism layer, a fully connected layer, and a regression layer connected in sequence. These layers form a complete network structure through a specific connection method, which is suitable for processing time series data such as this exemplary embodiment. It is also possible to prevent the model from overfitting and improve the generalization ability of the model by selecting a different network structure or adding a regularization term.

[0049] Optionally, the automatic optimization algorithm can be a swarm intelligence optimization algorithm that simulates automatic hunting behavior, which has the advantages of few parameters, strong adaptability, and easy implementation. For example, the GWO gray wolf optimization algorithm can be used to optimize the hyperparameters of the deep learning network to improve the efficiency of model training and the accuracy of prediction.

[0050] S3. Use the training set to train a water environment indicator change trend prediction model, wherein the basic network model of the water environment indicator change trend prediction model is a combination of a convolutional neural network, a long short-term memory neural network and a self-attention mechanism.

[0051] Optionally, the water environment index change trend prediction model can be trained using an Adam optimizer or an SGD optimization algorithm and corresponding first training parameters to achieve the best prediction effect. Optionally, the weight of the water environment index change trend prediction model in training can be optimized by a back propagation algorithm until a predetermined number of training iterations or convergence conditions are reached.

[0052] Optionally, the error index, determination coefficient or correlation coefficient of the water environment index change trend prediction model can be calculated through a pre-established test set to evaluate the performance of the water environment index change trend prediction model. Among them, the error index may include the sum of squared errors SSE, mean absolute error MAE, mean square error MSE, root mean square error RMSE and average percentage error MAPE. The correlation coefficient may be the Nash efficiency coefficient (Nash-Sutcliffe Efficiency Coefficient, referred to as NSE), which is an indicator used to evaluate the quality of the hydrological model simulation results. It quantifies the prediction accuracy of the model by comparing the simulated values ​​of the model with the actual observed values. The value range of NSE is -∞ to 1. The closer the value is to 1, the better the simulation effect of the model, that is, the higher the prediction accuracy of the model. If the NSE value is close to 0, it means that the prediction effect of the model is only equivalent to the average level of the observed values. If the NSE value is less than 0, it means that the prediction effect of the model is worse than random prediction, that is, the model is unreliable. The calculation formula of the Nash efficiency coefficient can be:

[0053]

[0054] Where NSE is the Nash efficiency coefficient; Q o,t is the observed value at the tth time point; Q m,t is the simulation value at the tth time point; is the average value of the observations; N is the total number of observations.

[0055] For example, the indicator change trend prediction results of the test set can be denormalized to obtain the actual indicator value to calculate the error between the prediction result and the true value, or the determination coefficient (R 2 ) and correlation coefficient to judge the predictive ability and generalization ability of the model.

[0056] Exemplary Embodiment 2

[0057] In this exemplary embodiment, the water environment index change trend prediction model can be obtained according to the water environment index change trend prediction model establishment method as described in the exemplary embodiment 1 above.

[0058] Exemplary Embodiment 3

[0059] In this exemplary embodiment, the method for predicting a water environment index change trend may include: using a water environment index change trend prediction model as in exemplary embodiment 2 to predict a water environment index change trend of a target water environment.

[0060] Optionally, the prediction results can be displayed using the response curve, loss iteration change curve, RMSE iteration change curve, feature graph, error graph and prediction effect graph generated by the water environment indicator change trend prediction model, providing researchers and decision makers with an intuitive display of model training and prediction effects.

[0061] Example 4

[0062] This exemplary embodiment provides a computer device. The computer device includes a processor and a memory. The memory is used to store a computer program. The computer program is executed by the processor so that the processor executes the computer program according to at least one of the water environment indicator change trend prediction model establishment method as in exemplary embodiment 1 and the water environment indicator change trend prediction method as in exemplary embodiment 3.

[0063] Exemplary Embodiment 5

[0064] The present exemplary embodiment provides a computer-readable storage medium storing a computer program. The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to execute at least one of the method for establishing a water environment indicator change trend prediction model as in exemplary embodiment 1 and the method for predicting water environment indicator change trends as in exemplary embodiment 3. The computer-readable recording medium is any data storage device that can store data read out by a computer system. Examples of computer-readable recording media include: read-only memory, random access memory, read-only optical disk, magnetic tape, floppy disk, optical data storage device, and carrier wave (such as data transmission through the Internet via a wired or wireless transmission path).

[0065] In order to better understand the above exemplary embodiments of the present invention, they are further described below in conjunction with specific embodiments and drawings, but the embodiments cited are not intended to limit the present invention.

[0066] Example 1

[0067] In this embodiment, the water environment index prediction is implemented by the following steps:

[0068] Step 1: Data preprocessing and analysis

[0069] Import the training data set, including water environment indicator data and its driving factors, and store it in the "data_ori" variable. Use the "clc" and "clear" commands to clear the memory and screen and initialize the computing environment. Ensure the repeatability of the results through "rng(0)". Also, determine the prediction step size "step" and the total data size "daynum", where "step=10".

[0070] Step 2: Data normalization

[0071] The feature input data "Features" and the actual value output data "W_data" are normalized, and the data is mapped to the [0,1] interval using the "mapminmax" function. The normalized data is stored in the "features" and "w_data" variables respectively.

[0072] Step 3: Dataset format conversion and division

[0073] Convert the normalized data into a format suitable for deep learning models, including input data "InputData" and output data "OutData". Determine the data window based on "NumDays" and "num_skip", and divide the data into training and test sets. Define the training set size as "num_train" and the test set size as "num_test_end".

[0074] Step 4: Automatic optimization of hyperparameters

[0075] Set the parameters of the automatic optimization algorithm, including the number of search agents "SearchAgents_no", the maximum number of iterations "Max_iter", the lower bound "lb" and the upper bound "ub" of the variable. Use the automatic optimization algorithm to optimize the hyperparameters of the deep learning network, including the learning rate, convolution kernel size, and the number of neurons. And store the optimization results in "Best_Pos" and "Best_Cost".

[0076] Step 5: Deep learning network construction and training

[0077] Construct a deep learning network structure, including convolutional layers, LSTM layers, and self-attention mechanism layers. Use the optimized hyperparameters to set the network training options, such as learning rate "learning_rate", convolution kernel size "KerlSize", and number of neurons "NumNeurons". Train the network and adjust the network weights through the back-propagation algorithm until the stop condition is met. The training process is shown in Table 1 below.

[0078] Table 1 Parameter automatic optimization neural network model training process

[0079]

[0080] Step 6: Prediction and performance evaluation

[0081] Use the trained network to predict the trend of water environment indicators in the test set, and store the prediction results in "bestPred". Denormalize the prediction results, such as using the "mapminmax('reverse',...)" function. Also, calculate the prediction error, including indicators such as SSE, MAE, MSE, RMSE and MAPE.

[0082] Step 7: Results display and analysis

[0083] Draw a comparison chart between the predicted results of negative water environment index change trends and the actual water environment index data, such as Figure 2 As shown. This can be achieved using commands such as "plot" and "figure". It can also display the loss iteration change curve and RMSE iteration change curve during the model training process.

[0084] In addition, the following aspects can be adjusted according to the specific application scenarios and data characteristics: 1) Data preprocessing methods to adapt to different data noise and outliers. 2) Normalization methods to optimize the model's adaptability to data distribution. 3) Network structure and layer configuration to adapt to different prediction tasks and data characteristics. 4) Automatic optimization algorithm parameter settings to improve model training efficiency and prediction accuracy. 5) Performance evaluation indicators to comprehensively evaluate the model's prediction performance.

[0085] Although the present invention has been described above in conjunction with the exemplary embodiments and the accompanying drawings, it should be apparent to those skilled in the art that various modifications may be made to the above-described embodiments without departing from the spirit and scope of the claims.

Claims

1. A method for establishing a water environment index change trend prediction model, characterized in that: The water environment index prediction model establishment method comprises the following steps: S1. Obtain historical indicator data of the target water environment and driving factor data of the historical indicator data for establishing a training data set, wherein the driving factor data includes meteorological data, hydrological data, pollution emission data from living sources, industrial pollution emission data, agricultural pollution emission data, endogenous pollution emission data and water body self-purification capacity data; S2. Using an automatic optimization algorithm to confirm the hyperparameters of the water environment index change trend prediction model, wherein the key hyperparameters include learning rate, convolution kernel size and number of neurons; S3. Using the training data set to train the water environment index change trend prediction model, wherein the basic network model of the water environment index change trend prediction model is a combination of a convolutional neural network, a long short-term memory neural network and a self-attention mechanism.

2. The method for establishing a water environment index change trend prediction model according to claim 1, characterized in that: The network structure of the water environment indicator change trend prediction model is a sequence input layer, a convolution layer, a ReLU activation layer, a sequence anti-folding layer, an LSTM layer, a self-attention mechanism layer, a fully connected layer and a regression layer connected in sequence.

3. The method for establishing a water environment index change trend prediction model according to claim 1, characterized in that: The automatic optimization algorithm is a swarm intelligence optimization algorithm that simulates automatic hunting behavior.

4. The method for establishing a water environment index change trend prediction model according to claim 1, characterized in that: In step S3, the water environment indicator change trend prediction model is trained using an Adam optimizer or an SGD optimization algorithm and corresponding first training parameters.

5. The method for establishing a water environment index change trend prediction model according to claim 1, characterized in that: In step S3, the weight of the water environment index change trend prediction model in training is optimized by a back propagation algorithm.

6. The method for establishing a water environment index change trend prediction model according to claim 1, characterized in that: In step S3, the error index, determination coefficient or correlation coefficient of the water environment index change trend prediction model is calculated by using a pre-established test set to evaluate the performance of the water environment index change trend prediction model.

7. A water environment index change trend prediction model, characterized in that: The water environment index change trend prediction model is obtained according to the water environment index change trend prediction model establishment method according to any one of claims 1 to 6.

8. A method for predicting the changing trend of water environment indicators, characterized in that: The method for predicting the changing trend of water environment indicators comprises: using the water environment indicator changing trend prediction model described in claim 7 to predict the changing trend of water environment indicators of the target water environment.

9. A computer device, characterized in that: The computer device comprises: at least one processor; and A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions include instructions for executing the method for establishing a water environment indicator change trend prediction model according to any one of claims 1 to 6, or instructions for executing the method for predicting the water environment indicator change trend according to claim 8.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method for establishing a water environment indicator change trend prediction model according to any one of claims 1 to 6 is implemented, or the method for predicting a water environment indicator change trend according to claim 8 is implemented.

Images