Underground water level prediction method and system based on space-time diagram convolutional network model

Abstract

The invention relates to the technical field of groundwater level prediction, in particular to a groundwater level prediction method and system based on a space-time diagram convolutional network model, and the method comprises the steps: obtaining groundwater level data, and extracting the time sequence features and labels of the groundwater level data; carrying out standardization processing on the extracted time sequence features, and creating a graph data object; defining a space-time convolutional network model layer structure, and constructing a space-time convolutional network model; the time sequence features and labels of the underground water level data and the graph data objects are combined into a water level data set, and the space-time convolutional network model is trained and optimized; the groundwater level is predicted through the optimized space-time convolutional network model, and water resources are managed according to the prediction result. According to the method, the problem that hidden space and time correlation in the spatio-temporal data is difficult to effectively capture is effectively solved, high-precision prediction of the underground water level is realized, and more scientific decision support is provided for water resource management.

Description

Technical Field

[0001] The present invention relates to the technical field of groundwater level prediction, and particularly to a groundwater level prediction method and system based on a spatio-temporal graph convolutional network model. Background Art

[0002] Groundwater level prediction is a complex process involving multiple spatio-temporal factors and their interactions, such as climate, terrain, hydrogeology, and time information, which makes groundwater level prediction challenging. Existing groundwater level prediction methods can be mainly divided into two categories: physics-based models and data-driven machine learning and deep learning models; traditional physical models rely on hydrodynamic mechanisms and water balance principles, closely combine with actual physical processes, and have strong physical interpretability. However, they require a large amount of hydrological, geological, and meteorological data as input and demand high-precision calibration of physical parameters, making it difficult to obtain data.

[0003] In contrast, data-driven models focus on directly learning the characteristics of spatio-temporal data from historical data. Especially models using deep learning techniques, such as convolutional neural networks, long short-term memory networks, and temporal convolutional networks, have strong adaptability and low computational costs. Through iterative calculations and model optimization, data-driven models can establish a mapping relationship between historical information and groundwater levels, and can better capture the spatio-temporal characteristics of groundwater levels and their non-linear relationships with other variables.

[0004] Currently, deep learning models have shown good performance in groundwater level prediction, especially when dealing with large-scale and high-dimensional spatio-temporal data. For example, LSTM combined with Dropout technology is used to improve the learning ability of the model, providing a groundwater level prediction model for certain regions. However, existing studies usually ignore the spatial dependence of time series data and only focus on the information in the time series, which may lead to inaccurate results in water level prediction and make it difficult to effectively capture the spatial and temporal correlations hidden in spatio-temporal data. In addition, traditional models often perform poorly when dealing with the non-local connectivity of spatio-temporal data. To overcome these deficiencies, the present invention proposes a groundwater level prediction method based on a spatio-temporal graph convolutional network, which simultaneously processes the temporal information and spatial information of water level data and adapts to complex spatio-temporal data. Summary of the Invention

[0005] The present invention provides a groundwater level prediction method and system based on a spatio-temporal graph convolutional network model, which can effectively solve the problems in the background art.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A groundwater level prediction method based on a spatio-temporal graph convolutional network model, the method comprising:

[0008] Obtain groundwater level data, and extract the temporal features and labels of the groundwater level data;

[0009] Perform standardization processing on the extracted temporal features, and create a graph data object according to the temporal features;

[0010] Define the layer structure of the spatio-temporal convolutional network model, and construct the spatio-temporal convolutional network model;

[0011] Merge the temporal features and labels of the groundwater level data and the graph data object into a water level data set, and use the water level data set to train and optimize the spatio-temporal convolutional network model;

[0012] Predict the groundwater level through the optimized spatio-temporal convolutional network model, and manage water resources according to the prediction results.

[0013] Further, the obtaining of the groundwater level data and the extraction of the temporal features and labels of the groundwater level data include:

[0014] Load the groundwater level data file, determine the study area, and select the observation well data of the study area;

[0015] Determine the sliding window length, and use the sliding window method to create temporal features and labels for each time point of the observation well data.

[0016] Further, the creating of the graph data object includes:

[0017] Take each observation well in the study area as a graph node to obtain a node feature matrix;

[0018] Set the correlation coefficient threshold, calculate the correlation coefficient of the water level changes between two observation wells, and if the water level change correlation coefficient is greater than the correlation coefficient threshold, establish an edge between the two observation wells to obtain the adjacency matrix of the graph;

[0019] Create a graph data object, and encapsulate the node feature matrix and the adjacency matrix.

[0020] Further, the constructed spatio-temporal convolutional network model includes:

[0021] Input layer, which receives the features of the groundwater level data of the water level data set;

[0022] Graph convolutional network layer, construct multiple graph convolutional network layers to extract the spatial features of the groundwater level data;

[0023] Graph attention network layer, which assigns different weights to neighboring observation wells through the attention mechanism;

[0024] A time feature extraction layer for extracting the time features of the groundwater level data, using a recurrent neural network as the time feature extraction layer;

[0025] An output layer that maps the extracted spatio-temporal features to the predicted groundwater level through a linear layer.

[0026] Furthermore, it includes: using a temporal convolutional network as the time feature extraction layer, and extracting the time dependence of the groundwater level data through multi-layer dilated convolution and residual connection.

[0027] Furthermore, it includes that when performing graph convolution calculation in the graph convolution network layer, introducing a graph convolution operator based on the spectral graph convolution method, adopting the Chebyshev polynomial approximation strategy, and setting an inter-layer multi-order dynamic selection mechanism according to the Chebyshev polynomial to optimize the graph convolution calculation.

[0028] Furthermore, it also includes that when performing graph convolution calculation in the graph convolution network layer, introducing a graph convolution operator based on the spectral graph convolution method, adopting a first-order approximation strategy to reduce the complexity of the graph convolution calculation.

[0029] Furthermore, using the water level dataset to train and optimize the spatio-temporal convolution network model, including:

[0030] Dividing the water level dataset into a training set, a validation set, and a test set;

[0031] Defining a loss function and an optimizer, and choosing the mean squared error loss function and the Adam optimizer;

[0032] Based on the training set and the validation set, iteratively training the spatio-temporal convolution network model through backpropagation and gradient descent algorithms;

[0033] Using the test set to evaluate the spatio-temporal convolution network model, calculating evaluation metrics, and tuning the model according to the evaluation results.

[0034] A groundwater level prediction system based on a spatio-temporal graph convolution network model, the system includes:

[0035] A time series feature acquisition module that acquires groundwater level data and extracts the time series features and labels of the groundwater level data;

[0036] A graph data object creation module that normalizes the extracted time series features and creates a graph data object according to the time series features;

[0037] A network model construction module that defines the layer structure of the spatio-temporal convolution network model and constructs the spatio-temporal convolution network model;

[0038] The network model training module combines the temporal features and labels of the groundwater level data and the graph data object into a water level dataset, and uses the water level dataset to train and optimize the spatio-temporal convolutional network model;

[0039] The groundwater level prediction module predicts the groundwater level through the optimized spatio-temporal convolutional network model, and manages water resources according to the prediction results.

[0040] Further, the graph data object creation module includes:

[0041] The observation well node establishment unit takes each observation well in the study area as a graph node to obtain a node feature matrix;

[0042] The observation well edge establishment unit sets a correlation coefficient threshold, calculates the water level change correlation coefficient between two observation wells, and if the water level change correlation coefficient is greater than the correlation coefficient threshold, establishes an edge between the two observation wells to obtain the adjacency matrix of the graph;

[0043] The matrix encapsulation unit creates a graph data object and encapsulates the node feature matrix and the adjacency matrix.

[0044] Through the technical solution of the present invention, the following technical effects can be achieved:

[0045] It effectively solves the problem of difficultly capturing the spatial and temporal correlations hidden in spatio-temporal data, realizes high-precision prediction of groundwater level. The groundwater level prediction method based on the spatio-temporal graph convolutional network model proposed by the present invention can not only cope with the spatio-temporal complexity of groundwater level prediction, provide accurate prediction results, but also provide more scientific decision-making support for water resource management.

[0046] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the specific embodiments of the present application are hereinafter specifically exemplified. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0048] Figure 1 It is a schematic flowchart of a groundwater level prediction method based on a spatio-temporal graph convolutional network model;

[0049] Figure 2 It is a schematic structural diagram of an observation well network;

[0050] Figure 3 It is a schematic structural diagram of a groundwater level prediction system based on a spatio-temporal graph convolutional network model. Specific implementation manners

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0053] Embodiment 1

[0054] As Figure 1 shown, a groundwater level prediction method based on a spatio-temporal graph convolutional network model, the method includes:

[0055] S1: Obtain groundwater level data, and extract the temporal features and labels of the groundwater level data;

[0056] S2: Perform standardization processing on the extracted temporal features, and create a graph data object according to the temporal features;

[0057] Specifically, in order to ensure that the original data can be constructed into a graph structure with both temporal and spatial information after reasonable preprocessing, and provide high-quality input for subsequent model construction, it is necessary to ensure that the obtained data has temporality and regional representativeness, and perform preprocessing on the data; extract the key temporal change laws from the original data, and at the same time determine the prediction target (label) to provide basic information for subsequent model training; and perform normalization or standardization on the extracted temporal features to eliminate the influence of different dimensions or data scales, make the data more stable, and facilitate the rapid convergence of the model; as a key step to extend the temporal data to the spatial domain, a graph data object can be created to provide necessary spatial dependence information for subsequent graph convolution operations.

[0058] S3: Define the layer structure of the spatio-temporal convolutional network model, and construct the spatio-temporal convolutional network model;

[0059] S4: Combine the temporal features, labels of the groundwater level data, and the graph data object into a water level dataset, and use the water level dataset to train and optimize the spatio-temporal convolutional network model;

[0060] S5: Predict the groundwater level through the optimized spatio-temporal convolutional network model, and manage the water resources according to the prediction results.

[0061] Specifically, to ensure that the model can effectively extract the complex spatio-temporal relationships in the groundwater level data, the designed model structure can not only capture the temporal dynamic changes but also learn the spatial relationships; the collected data can be used to train the spatio-temporal convolutional network, and the network parameters can be adjusted through the optimization algorithm to enable the model to accurately capture the spatio-temporal patterns in the data and improve the prediction accuracy; deploy the optimized spatio-temporal convolutional network model into practice to predict the groundwater level. By accurately predicting the change trend of the groundwater level, it can provide more accurate decision-making support for water resource management. Relevant departments can take corresponding prevention and control measures in advance, allocate water resources reasonably, and ensure key water conservancy needs such as agricultural irrigation and urban water supply; in addition, this method can also be extended and applied to the prediction of other hydrological fields, such as groundwater level, river water level, etc., with broad application prospects and important social value.

[0062] Through the technical solution of the present invention, the problem of difficultly effectively capturing the spatial and temporal correlations hidden in the spatio-temporal data is effectively solved, and high-precision prediction of the groundwater level is realized. The groundwater level prediction method based on the spatio-temporal graph convolutional network model proposed by the present invention can not only cope with the spatio-temporal complexity of groundwater level prediction, provide accurate prediction results, but also provide more scientific decision-making support for water resource management.

[0063] As a preference of this embodiment, obtaining the groundwater level data and extracting the temporal features and labels of the groundwater level data includes:

[0064] S11: Load the groundwater level data file, determine the study area, and select the observation well data of the study area;

[0065] S12: Determine the sliding window length, and use the sliding window method to create temporal features and labels for each time point of the observation well data.

[0066] In this embodiment, groundwater level data files can be loaded from a specified folder, and these data files can be read and merged using the pandas library. The locations of the observation wells within the study area are selected, and the groundwater level data for this area is extracted. To better utilize the spatio-temporal characteristics of the data, a sliding window method can be used to construct features and labels. Specifically, for each time point, the water level data for the previous N days can be selected as features, and the water level value at the current time point can be used as the label. Through normalization processing, the water level data is converted into a form suitable for model training to remove the dimensional differences in the data and improve the stability and accuracy of model training.

[0067] Preferably, in this embodiment, creating a graph data object includes:

[0068] S21: Taking each observation well in the study area as a graph node to obtain a node feature matrix;

[0069] S22: Setting a correlation coefficient threshold, calculating the correlation coefficient of the water level changes between two observation wells. If the correlation coefficient of the water level changes is greater than the correlation coefficient threshold, an edge is established between the two observation wells to obtain the adjacency matrix of the graph;

[0070] S23: Creating a graph data object and encapsulating the node feature matrix and the adjacency matrix.

[0071] Specifically, in the graph structure construction stage, the locations of the observation wells within the study area can be regarded as the nodes of the graph, and the spatial correlation between the observation wells is used to construct the edges of the graph. Specifically, the correlation of the water level changes between every two observation wells is calculated (such as distance-based correlation or time series correlation of the water level). If the correlation coefficient exceeds the set threshold, an edge is established between these two observation wells. In this way, a graph structure reflecting the spatial correlation between the observation wells can be obtained, as Figure 2 shown. Assuming the number of nodes in the graph is N, the set of edges is E, and the node feature matrix is X, then the graph can be represented as G=(V, E), where V is the set of observation well nodes, and E is the set of edges representing the connection relationship between the observation wells. By constructing such a graph structure, the spatial correlation between the observation wells can be fully utilized to provide data input for the spatio-temporal graph convolutional network model.

[0072] Furthermore, the constructed spatio-temporal convolutional network model includes:

[0073] An input layer that receives the features of the groundwater level data in the water level data set;

[0074] A graph convolutional network layer that constructs multiple layers of graph convolutional network layers to extract the spatial features of the groundwater level data;

[0075] A graph attention network layer that assigns different weights to neighboring observation wells through an attention mechanism;

[0076] In this embodiment, a graph convolutional network and a graph attention network are used to extract the spatial features of groundwater level data. The graph convolutional network aggregates the spatial features of the observation wells with the features of their neighboring observation wells through convolutional operations, so as to obtain a new feature representation of the observation wells. Let H (l) be the feature representation of the observation wells at the l-th layer, A be the adjacency matrix of the graph, and W (1) be the weight matrix at the -th layer. Then, the output of the graph convolutional network at the l-th layer can be expressed as:

[0077] H (1+1) = σ(AH (1) W (1) );

[0078] where σ is the activation function; the graph convolutional layer propagates the spatial information between the observation wells through the adjacency matrix, so as to capture the correlation between the observation wells and obtain the spatial dependence relationship of the groundwater level.

[0079] The graph attention network further introduces an adaptive attention mechanism, which aggregates the node features by assigning different weights to each neighboring observation well, so as to optimize the representation ability of the spatial features. Let a be the parameter of the attention mechanism, and a ij be the attention weight between node i and node j. Then, the output of the graph attention network can be expressed as:

[0080] H i (l+1) = σ(∑ j∈N(i) a ij H j (l) W (l) );

[0081] where N(i) is the set of neighboring observation wells of node i; the graph attention network can not only automatically learn the importance of the neighboring observation wells, enhance the spatial feature representation, but also effectively capture the spatial dependence relationship of the groundwater level.

[0082] The time feature extraction layer is used to extract the time features of the groundwater level data, and a recurrent neural network is used as the time feature extraction layer; the recurrent neural network includes long short-term memory networks and gated recurrent units, which can effectively capture the long-term time dependence of the groundwater level.

[0083] The output layer maps the extracted spatio-temporal features to the predicted groundwater level through a linear layer, converts the high-dimensional spatial and temporal feature spaces into groundwater level prediction values, so as to ensure that the model output can be used for practical applications.

[0084] The present invention uses a spatio-temporal graph convolutional network model for groundwater level prediction, which mainly consists of spatio-temporal convolutional blocks, combining graph convolutional layers and gated temporal convolutional layers to make full use of the spatio-temporal characteristics of groundwater level data. The core idea of the spatio-temporal graph convolutional network model is to extract the spatial features of the monitoring data of observation wells through graph convolutional operations and capture the long-range dependencies in the time series through a time convolutional network, so as to achieve more accurate water level prediction.

[0085] On the basis of the above embodiments, a time convolutional network is used as the time feature extraction layer, and through multi-layer dilated convolution and residual connections, the time dependencies of groundwater level data are extracted.

[0086] Specifically, the above embodiments use a cyclic structure to process time series data and gradually transfer hidden states; while this embodiment uses a convolutional structure to process time series, expands the receptive field through multi-layer dilated convolution, and at the same time introduces residual connections, which is beneficial to capturing long-term dependencies and solving the training problems of deep networks, has high computational efficiency, and can be calculated in parallel, avoiding the problems of gradient disappearance and explosion that may occur in traditional RNNs.

[0087] Furthermore, when performing graph convolution calculations in the graph convolutional network layer, a graph convolution operator is introduced based on the spectral graph convolution method, the Chebyshev polynomial approximation strategy is adopted, and a multi-order dynamic selection mechanism between layers is set according to the Chebyshev polynomial to optimize the graph convolution calculation.

[0088] As a preference of this embodiment, the spatio-temporal graph convolutional network model can extract the spatial features of groundwater level data through graph convolutional layers. Since the distribution of observation wells usually presents a graph structure, traditional methods ignore this spatial attribute when processing water level data, while this model directly applies graph convolution to graph-structured data and can effectively capture the spatial correlation between observation wells. When calculating graph convolution, a graph convolution operator is introduced based on the concept of spectral graph convolution, but due to its high original computational cost, the Chebyshev polynomial approximation strategy can be adopted to reduce the computational complexity. The Chebyshev polynomial approximation rewrites the graph convolution as:

[0089]

[0090] where is the k-th order Chebyshev polynomial, θ k is the coefficient to be learned, K is the order of the polynomial, is the normalized Laplacian matrix, and x is the input feature.

[0091] Using Chebyshev polynomial approximation can reduce the cost of calculating the budget groundwater level, but only capture the low-order feature information within the local neighborhood, while the high-order information may be ignored. Therefore, a multi-order dynamic selection mechanism between layers can be set up to obtain richer information. Define multiple parallel branches in the graph convolutional network layer, each branch corresponding to a Chebyshev polynomial approximation of a different order, and set up a dynamic selection mechanism to determine the most appropriate order. In this way, both global information can be captured and the computational efficiency can be maintained to meet the requirements of different tasks.

[0092] Furthermore, when performing graph convolution calculations in the graph convolutional network layer, introduce a graph convolution operator based on the spectral graph convolution method and adopt a first-order approximation strategy to reduce the complexity of graph convolution calculations.

[0093] Based on the above embodiments, a first-order approximation strategy can also be used to reduce the complexity of graph convolution calculations. The first-order approximation constructs a deeper architecture by stacking multiple local graph convolutional layers using the first-order approximation of the graph Laplacian matrix.

[0094] As a preference of this embodiment, use the water level dataset to train and optimize the spatio-temporal convolutional network model, including:

[0095] S41: Divide the water level dataset into a training set, a validation set, and a test set;

[0096] S42: Define a loss function and an optimizer, and select the mean squared error loss function and the Adam optimizer;

[0097] S43: Based on the training set and the validation set, iteratively train the spatio-temporal convolutional network model through backpropagation and the gradient descent algorithm;

[0098] S44: Use the test set to evaluate the spatio-temporal convolutional network model, calculate the evaluation metrics, and perform model tuning according to the evaluation results.

[0099] Specifically, for model training and optimization, it can be divided into a training set, a validation set, and a test set according to a certain ratio. For example, use 80% of the data as the training set, 10% as the validation set, and 10% as the test set to ensure the generalization ability of the model. When training, the mean squared error (MSE) can be used as the loss function to measure the difference between the predicted value and the actual value, and the Adam optimizer is used for training to ensure efficient convergence and less computational overhead. Let be the actual value, be the predicted value, then the calculation formula of MSE is:

[0100]

[0101] During the training process, the model is iteratively trained through backpropagation and gradient descent algorithms. Each training step includes forward propagation, calculating the loss, backpropagation, and updating the weights. Through continuous iterative optimization, the optimal model parameters are gradually found, and the hyperparameters are optimized based on the performance of the validation set to avoid overfitting and improve the prediction ability of the model. The test set is used to evaluate the model, and evaluation metrics such as mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) are calculated. These metrics help to comprehensively evaluate the prediction performance of the model, and the model is tuned according to the evaluation results.

[0102] Through the above model training, a spatio-temporal graph convolutional network model capable of accurately predicting groundwater levels can be obtained. In practical applications, when the latest groundwater level data is input, the model can predict future water level changes based on historical data and spatial correlations. This method can not only be used for predicting groundwater levels but also be extended to other water resource monitoring systems, such as predicting tasks for groundwater levels, river water levels, etc. By combining spatio-temporal information, the model can fully capture the spatial distribution characteristics and time dependencies in the hydrological process, thus providing more accurate prediction results.

[0103] Embodiment 2:

[0104] As Figure 3 shown, a groundwater level prediction system based on a spatio-temporal graph convolutional network model, the system includes:

[0105] A time series feature acquisition module, which acquires groundwater level data and extracts the time series features and labels of the groundwater level data;

[0106] A graph data object creation module, which standardizes the extracted time series features and creates a graph data object based on the time series features;

[0107] A network model construction module, which defines the layer structure of the spatio-temporal convolutional network model and constructs the spatio-temporal convolutional network model;

[0108] A network model training module, which combines the time series features and labels of the groundwater level data and the graph data object into a water level data set, and uses the water level data set to train and optimize the spatio-temporal convolutional network model;

[0109] A groundwater level prediction module, which predicts the groundwater level through the optimized spatio-temporal convolutional network model and manages the water resources according to the prediction results.

[0110] The above adjustment system in the present invention can effectively implement the groundwater level prediction method based on the spatio-temporal graph convolutional network model, and the technical effects that can be achieved are as described in the above embodiments, which will not be elaborated here.

[0111] Furthermore, the graph data object creation module includes:

[0112] An observation well node establishment unit takes each observation well in the study area as a graph node to obtain a node feature matrix;

[0113] An observation well edge establishment unit sets a correlation coefficient threshold, calculates the correlation coefficient of the water level changes between two observation wells. If the correlation coefficient of the water level changes is greater than the correlation coefficient threshold, an edge is established between the two observation wells to obtain the adjacency matrix of the graph;

[0114] A matrix encapsulation unit creates a graph data object and encapsulates the node feature matrix and the adjacency matrix.

[0115] Similarly, for the above optimization solutions of the system, the corresponding optimization effects of the methods in Embodiment 1 can also be respectively achieved, and will not be elaborated here again.

[0116] Although the present application has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, the present specification and the drawings are merely exemplary descriptions of the present application defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A groundwater level prediction method based on a spatiotemporal graph convolutional network model, characterized in that: The method comprises: Acquire groundwater level data, and extract time series features and labels of the groundwater level data; Standardizing the extracted time series features and creating a graph data object according to the time series features; Defining a spatiotemporal convolutional network model layer structure and constructing the spatiotemporal convolutional network model; Merging the time series features and labels of the groundwater level data and the graph data object into a water level dataset, and using the water level dataset to train and optimize the spatiotemporal convolutional network model; The groundwater level is predicted by the optimized spatiotemporal convolutional network model, and water resources are managed according to the prediction results.

2. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 1 is characterized in that: The step of obtaining groundwater level data and extracting time series features and labels of the groundwater level data includes: Load the groundwater level data file, determine the study area, and select the observation well data of the study area; The sliding window length is determined, and a sliding window method is used to create time series features and labels for each time point of the observation well data.

3. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 1 is characterized in that: The creating of the graph data object comprises: Each observation well in the study area is taken as a graph node to obtain the node feature matrix; A correlation coefficient threshold is set, and the correlation coefficient of water level changes of two observation wells is calculated. If the correlation coefficient of water level changes is greater than the correlation coefficient threshold, an edge is established between the two observation wells to obtain an adjacency matrix of the graph; A graph data object is created to encapsulate the node feature matrix and the adjacency matrix.

4. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 1 is characterized in that: The constructed spatiotemporal convolutional network model includes: An input layer, receiving the characteristics of groundwater level data of the water level dataset; A graph convolutional network layer, constructing a multi-layer graph convolutional network layer to extract the spatial characteristics of the groundwater level data; Figure attention network layer, which assigns different weights to neighboring observation wells through the attention mechanism; A time feature extraction layer, used to extract the time features of the groundwater level data, using a recurrent neural network as the time feature extraction layer; The output layer maps the extracted spatiotemporal features to the predicted groundwater level through a linear layer.

5. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 4 is characterized in that: include: A temporal convolutional network is used as the temporal feature extraction layer, and the temporal dependency of the groundwater level data is extracted through multi-layer dilated convolution and residual connection.

6. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 4 is characterized in that: This includes introducing a graph convolution operator based on a spectral graph convolution method when performing graph convolution calculations in the graph convolution network layer, adopting a Chebyshev polynomial approximation strategy, and setting an inter-layer multi-order dynamic selection mechanism according to the Chebyshev polynomial to optimize the graph convolution calculations.

7. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 4 is characterized in that: It also includes introducing a graph convolution operator based on a spectral graph convolution method when the graph convolution network layer performs graph convolution calculations, and adopting a first-order approximation strategy to reduce the complexity of the graph convolution calculations.

8. The groundwater level prediction method based on the spatiotemporal graph convolutional network model according to claim 1 is characterized in that: Using the water level dataset to train and optimize the spatiotemporal convolutional network model includes: Dividing the water level data set into a training set, a validation set and a test set; Define the loss function and optimizer, select the mean square error loss function and Adam optimizer; Based on the training set and the validation set, iteratively training the spatiotemporal convolutional network model through back propagation and gradient descent algorithms; The test set is used to evaluate the spatiotemporal convolutional network model, calculate evaluation indicators, and perform model tuning based on the evaluation results.

9. A groundwater level prediction system based on a spatiotemporal graph convolutional network model, characterized in that: The system comprises: A time series feature acquisition module is used to acquire groundwater level data and extract time series features and labels of the groundwater level data; A graph data object creation module, which performs standardization processing on the extracted time series features and creates a graph data object according to the time series features; A network model construction module defines a spatiotemporal convolutional network model layer structure and constructs the spatiotemporal convolutional network model; A network model training module, which combines the time series features and labels of the groundwater level data and the graph data object into a water level data set, and uses the water level data set to train and optimize the spatiotemporal convolutional network model; The groundwater level prediction module predicts the groundwater level through the optimized spatiotemporal convolutional network model and manages water resources according to the prediction results.

10. The groundwater level prediction system based on the spatiotemporal graph convolutional network model according to claim 9 is characterized in that: The graph data object creation module includes: The observation well node establishment unit takes each observation well in the study area as a graph node and obtains the node feature matrix; An observation well edge establishment unit is provided, a correlation coefficient threshold is set, and a water level change correlation coefficient of two observation wells is calculated. If the water level change correlation coefficient is greater than the correlation coefficient threshold, an edge is established between the two observation wells to obtain an adjacency matrix of the graph; The matrix encapsulation unit creates a graph data object and encapsulates the node feature matrix and the adjacency matrix.

Images