CN114841402A - A method and system for predicting groundwater level height based on multi-feature map network - Google Patents

Abstract

The invention discloses a method and system for predicting groundwater level height based on a multi-feature graph network. The method includes: selecting a plurality of groundwater level monitoring points at different locations as spatial nodes for constructing a graph structure data set, and obtaining the characteristics of each node information, and construct the feature vector of each node; based on the feature vector, the correlation between two nodes is weighed, and the edge structure information used to describe the connectivity of the node is constructed, and the inverse of the geographical distance between the two connected nodes is used as the edge connection of the nodes. weights; build a graph structure data set based on the above information; improve the GCN to obtain a groundwater level prediction model; and train the groundwater level prediction model based on the graph structure data set; Prediction of groundwater levels. The invention can realize the efficient and accurate prediction of the groundwater level in the unknown position.

Description

A method and system for predicting groundwater level height based on multi-feature map network

technical field

The invention relates to the technical field of hydrological geography intelligent decision-making, in particular to a method and system for predicting groundwater level height based on a multi-feature map network.

Background technique

Surface water and groundwater resources are important components in the complex ecological water cycle system. The main water sources in the water resource system are atmospheric water, surface water and groundwater, as well as water transferred from outside the system. Various water sources can be transformed into each other under certain conditions. For example, rainfall infiltration and irrigation can replenish soil water. After the soil water is saturated, it continues to infiltrate to form groundwater. Among them, for irrigated agricultural areas, long-term river openings for irrigation and poor drainage of flood irrigation have led to high groundwater levels. In addition, surface evaporation in irrigated agricultural areas is usually relatively strong, resulting in the deepening of soil salinization and the gradual deterioration of the ecological environment. . With the gradual scarcity of surface water resources, the river runoff has been greatly reduced, and the amount of irrigation water has also decreased year by year.

The transformation relationship between surface water and groundwater is very complex, and the change of groundwater level is not only affected by the amount of surface irrigation, but also related to the topography. Studying the replenishment and discharge relationship between surface water and groundwater in different geographical locations can effectively alleviate the problem of soil salinization caused by the unreasonable use of water resources, and provide accurate guidance for agricultural production. Existing research mainly uses professional knowledge of hydrogeology to study the evolution relationship between surface water and groundwater, and simulates the scheduling process of water resources in the system by establishing traditional hydrogeological models and mathematical models. Although this method has a strong theoretical basis, it cannot analyze the precise evolution of water resources from a data perspective for complex water resources systems.

Therefore, an intelligent solution that comprehensively considers data characteristics and principle characteristics is required.

SUMMARY OF THE INVENTION

The present invention provides a method and system for predicting groundwater level height based on a multi-feature map network, so as to solve the technical problem of lack of quantitative and preciseness in the dynamic evolution process of existing surface water and groundwater.

In order to solve the above-mentioned technical problems, the present invention provides the following technical solutions:

In one aspect, the present invention provides a method for predicting groundwater level height based on a multi-feature map network, and the method for predicting groundwater level height based on a multi-feature map network includes:

Selecting a plurality of groundwater level monitoring points at different locations as spatial nodes for constructing a graph structure data set, acquiring characteristic information of each node, and constructing a characteristic vector of each node based on the characteristic information;

Weigh the correlation between the two nodes based on the feature vector, construct edge structure information for describing the connectivity of the nodes, and use the inverse of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

Using the feature information and edge structure information of the nodes as the sample features, and the depth interval category to which the groundwater level corresponding to the nodes belongs as the sample labels, a graph structure data set is constructed;

Improving a graph convolutional network (Graph Convolutional Network, GCN) to obtain a groundwater level prediction model; and training the groundwater level prediction model based on the graph-structured data set;

Based on the trained groundwater level prediction model, the prediction of the groundwater level of the spatial location is realized.

Further, the feature information includes surface water resources information and geospatial information;

The surface water resources information includes the irrigation amount, drainage amount, and total amount of precipitation in the preset period of the area where the node is located; the geospatial information includes: vadose zone lithology, landform type, permeability K1 partition, land type name , total dissolved solids, permeability coefficient KCP, and distance between nodes and branch channels.

Further, the feature vectors of each node are respectively constructed based on the feature information, including:

Perform vectorization processing on the feature information to obtain the feature vector of each node; wherein, the vectorized processing method for the feature information is one-hot encoding and normalization, and one-hot encoding is performed for non-quantized information first. , and then perform normalization processing. For quantized information, perform normalization processing directly.

Further, based on the feature vector, the correlation between the two nodes is weighed, and edge structure information for describing the connectivity of the nodes is constructed, including:

Calculate cosine similarity, Pearson correlation coefficient and Euclidean distance between feature vectors of all nodes;

Construct a one-way connected relationship graph of all nodes, retain the connected edges that meet the preset conditions, and regard the nodes corresponding to the connected edges that do not meet the preset conditions as non-connected relationships; wherein, the preset conditions are simultaneously satisfying the relationship between feature vectors. The cosine similarity is greater than 0.7, the Pearson correlation coefficient is greater than 0.8, and the Euclidean distance is less than 1;

The connectivity relationship retained after filtering is counted to obtain edge structure information used to describe the connectivity of nodes.

Further, the depth interval categories include 0-1m, 1-2m, 2-3m, 3-4m, 4-5m and more than 5m.

Further, improve the GCN, including:

The weight matrix is added to the graph convolution operation of the first and last layers of the GCN network. The specific implementation is as follows: the spatial node introduces the edge connection weight as the aggregation coefficient in the stage of aggregating neighbor node features.

Further, the GCN is improved, including:

A graph attention convolution module is added after the first graph convolution layer of the GCN network. The input feature of the graph attention convolution module is the feature output after the first layer graph convolution operation aggregates neighbor nodes. The output feature of the attention convolution module is the aggregated feature after automatically learning and updating the weight of the node and its neighbor nodes; after obtaining the feature output of the graph attention convolution module, it is compared with the output feature of the first graph convolution layer. Fusion, the fused features are fed into the last graph convolutional layer.

Further, after realizing the prediction of the groundwater level at the spatial location, the method further includes:

Evaluate the validity of the prediction results of the groundwater level prediction model.

Further, the effectiveness evaluation of the prediction results of the groundwater level prediction model includes:

Compare the categories of the predicted results with the real results, and evaluate the effectiveness of the model by the prediction accuracy.

On the other hand, the present invention also provides a groundwater level height prediction system based on a multi-feature map network, and the groundwater level height prediction system based on the multi-feature map network includes:

Graph-structured dataset building blocks for:

Selecting a plurality of groundwater level monitoring points at different locations as spatial nodes for constructing a graph structure data set, acquiring characteristic information of each node, and constructing a characteristic vector of each node based on the characteristic information;

Weigh the correlation between the two nodes based on the feature vector, construct edge structure information for describing the connectivity of the nodes, and use the inverse of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

Using the feature information and edge structure information of the nodes as the sample features, and the depth interval category to which the groundwater level corresponding to the nodes belongs as the sample labels, a graph structure data set is constructed;

The groundwater height prediction model construction and training module is used to improve the graph convolutional network (Graph Convolutional Network, GCN) to obtain the groundwater level prediction model; groundwater level prediction model for training;

The groundwater height prediction module is used to construct a groundwater level prediction model trained by the training module based on the groundwater height prediction model, so as to realize the prediction of the groundwater level of the spatial location.

In another aspect, the present invention also provides an electronic device, which includes a processor and a memory; wherein, the memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the above method.

In yet another aspect, the present invention also provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, and the instruction is loaded and executed by a processor to implement the above method.

The beneficial effects brought by the technical solution provided by the present invention at least include:

The groundwater level height prediction method provided by the present invention is a method for predicting the groundwater level height through surface water resources and geographic space attributes. The graph structure data set is constructed in units of spatial position nodes, the distance weight between nodes is added, and attention mechanism and feature fusion mechanism are introduced, so that this method can obtain better feature expression of nodes. At the same time, this method can be used in some nodes. In the case of , the semi-supervised learning method is used to train the model, so as to realize the accurate prediction of the groundwater level in the spatial location. The invention studies the evolution relationship between surface water and groundwater through data-oriented artificial intelligence technology, and solves the problems of inability to quantify and lack of accuracy in the existing surface water and groundwater evolution methods under complex conditions.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a schematic flowchart of the execution of a method for predicting groundwater level height based on a multi-feature map network provided by an embodiment of the present invention;

2 is a schematic diagram of a spatial node feature descriptor provided by an embodiment of the present invention;

3 is a flow chart of attention factor calculation provided by an embodiment of the present invention;

4 is a schematic diagram of an improved GCN network structure provided by an embodiment of the present invention;

5 is a schematic diagram of a confusion matrix of a prediction result of a test set space node provided by an embodiment of the present invention;

6 is a schematic structural diagram of a groundwater level height prediction system based on a multi-feature map network provided by an embodiment of the present invention;

Figure 7 is a block diagram of an electronic device to which the method of the present invention is applied.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

first embodiment

This embodiment provides a method for predicting groundwater level height based on a multi-feature map network, and the method can be implemented by an electronic device, and the electronic device can be a terminal or a server. The execution flow of the groundwater level height prediction method based on the multi-feature map network is shown in Figure 1, including the following steps:

S1, selecting multiple groundwater level monitoring points at different locations as spatial nodes for constructing a graph structure data set, acquiring characteristic information of each node, and constructing a characteristic vector of each node based on the characteristic information;

S2, weigh the correlation between two nodes based on the feature vector, construct edge structure information for describing the connectivity of the nodes, and use the inverse of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

S3, using the feature information and edge structure information of the node as the sample feature, and using the depth interval category to which the buried depth of the groundwater level corresponding to the node belongs as the sample label, construct a graph structure data set;

S4, improve the graph convolutional network (Graph Convolutional Network, GCN) to obtain a groundwater level prediction model; and train the groundwater level prediction model based on the graph structure data set;

S5, based on the trained groundwater level prediction model, realize the prediction of the groundwater level of the spatial location.

Among them, the feature information in S1 includes surface water resources information and geospatial information; surface water resources information includes irrigation volume, drainage volume, and total precipitation in the preset period of the area where the node is located; geospatial information includes: vadose zone rock properties, landform type, permeability K1 zone, land type name, total dissolved solids, permeability coefficient KCP, and the distance between nodes and branch channels. Based on this, the implementation process of S1 is as follows:

Obtain the surface irrigation and drainage, meteorological data and geographic attribute information of multiple locations in the space, and build a feature vector used to describe the surface information of the spatial location nodes based on the obtained information; among them, the surface irrigation and drainage data are flow statistics, and the meteorological data are height statistics. , the geographic attribute data is text information; the characteristics of spatial nodes are generated by establishing the node mapping between groundwater monitoring points and surface spatial locations, including: standardizing the surface water volume calculation for irrigation and drainage data and meteorological data and assigning them to the grid cells where the nodes are located. The geographic attribute information is feature encoded, and the node feature vector is obtained by fusion. The specific implementation process includes the following steps:

According to the irrigation canal system, the irrigation area is divided into several sub-basins. Each sub-basin of the irrigation area has clear water diversion sources and data statistics. The drainage source considers the entire agricultural irrigation area as a whole, and the drainage is carried out according to the weight of the corresponding sub-basin irrigated area. data distribution.

Divide the sub-watershed into grid units of 100m*100m, superimpose the divided grid units with the above-mentioned drainage sub-basins, establish the topological relationship between grid units and drainage units, and clarify the drainage unit where each grid unit is located. The calculation method of water resources characteristics of water area, drainage area and grid unit is as follows:

Daily irrigation and drainage volume = daily average irrigation and drainage flow × 24 × 60 × 60

Daily precipitation evaporation = daily precipitation evaporation height × sub-basin area × 666.7 × 10 ^-4

Establish a mapping relationship between groundwater monitoring points and surface grid cells to form spatial nodes. According to the coordinate position of the spatial node, its geographical and spatial attributes are counted, including: lithology of vadose zone, landform type, permeability K1 partition, land type name, total dissolved solids, permeability coefficient KCP, and the distance between nodes and branch canals. The different permeability values of vadose zone lithology include sand, clay sandy soil, interlayer water distribution area, upper clay soil and lower sandy soil, etc.; landform types include aeolian dunes, river-lake plains, alluvial-proluvial plains , modern alluvial fans, piedmont alluvial inclined plains, hills and platforms, etc.; the permeability K1 partition value range is 1-3, 3-5, 5-10, 10-20 corresponding to different permeability; the land type names include Corn, wheat, grapes, wolfberry, rice, villages, earth vegetables, cities, abandoned land, etc. Geospatial attributes are collected through the ArcGis spatial connection function. For polygon type layers, select the polygon into which the node falls, and for point type and line type layers, select the point or line closest to the node. After the collection is completed, the geospatial attributes and water resources characteristics are integrated to construct the node feature descriptor. The spatial node feature descriptor is shown in Figure 2.

After obtaining the feature descriptor (feature information) of each spatial node, vectorize the spatial node descriptor. The specific vectorization method is: one-hot encoding and normalization. For non-quantized columns, first encode and then normalize. , and the quantization column is normalized to obtain the feature vector of each spatial node.

Further, in this embodiment, the feature matrix of the spatial nodes is represented in the form of a two-dimensional matrix, wherein the row index represents the number and index of the spatial nodes, and the column index represents the attribute characteristics of the nodes, including surface irrigation and drainage, meteorological data and multi-dimensional terrain. Geomorphic attribute data.

Further, in this embodiment, in the above S2, the edge structure information is constructed by calculating specific indicators of the feature vector, including three indicators of cosine similarity, Pearson correlation coefficient, and Euclidean distance. Cosine similarity evaluates their similarity by calculating the cosine value of the angle between two vectors to determine whether the two vectors are roughly pointing in the same direction. The closer the cosine value is to 1, the higher the similarity. Calculated as follows:

The Pearson correlation coefficient is used to measure whether two vectors are on a line, and is used to measure the linear correlation between the distance variables. It is obtained by calculating the covariance and standard deviation between the two samples. The absolute value of the correlation coefficient The closer to 1, the stronger the linear correlation. Calculated as follows:

Euclidean distance is a commonly used definition of distance, which refers to the real distance between two points in m-dimensional space. It is used to find the distance between two vectors. The value ranges from 0 to positive infinity. Obviously, if two vectors The distance of the space is smaller, then the vectors must be more similar. The calculation formula is as follows:

Specifically, the construction process of the above-mentioned spatial node connectivity edge structure is as follows:

After obtaining the connected edge structure information, it corresponds to the geographic space according to the index, and obtains the geographic coordinates of the spatial nodes, calculates the geographic distance of the connected nodes through the geographic coordinates, and takes the reciprocal of the distance as the edge connection of the spatial nodes in the network model. The weight is output to an n×n sparse matrix, and the corresponding distance weight of nodes without edge connections is 0. In the process of graph convolution, neighbor nodes with a closer distance can aggregate more information, and neighbor nodes with a distance away contribute less feature information.

Further, in this embodiment, each correlation evaluation index and the weight of the connected edge are represented by a two-dimensional matrix, the row and column values represent the spatial node index, and the matrix elements are the correlation evaluation index between nodes and the calculation result of the reciprocal distance; connectivity; The edge structure information is represented by an n*2 matrix, each row of the matrix has two spatial node indices, indicating that there is connectivity between the two nodes.

Further, in this embodiment, the depth interval categories in the above S3 include 0-1m, 1-2m, 2-3m, 3-4m, 4-5m and more than 5m, corresponding to different interval categories; It is divided into training set, validation set and test set; among them, the training set is used to train the groundwater level height prediction model, the validation set is used to verify the accuracy of the model during the training process, and the test set is used to evaluate the trained model. Another semi-supervised learning method trains the model on nodes with known class labels to predict spatial nodes with unknown class labels. Specifically, the implementation process of the above S3 is as follows:

In the agricultural irrigation area, 200 groundwater level monitoring points were selected as the spatial nodes of the graph dataset constructed by this method. The attributes used to describe the characteristics of spatial nodes include surface water resources attributes and geospatial attributes. The attributes of surface water resources include monthly irrigation, drainage and precipitation of spatial grids, and the attributes of geographic space include vadose zone lithology, landform type, permeability K1 partition, land type name, total dissolved solids, permeability coefficient KCP, node and canal distance. The groundwater level depth of the groundwater monitoring point is used as the category label of the space node, and each category includes a certain groundwater level depth range. For example, burial depths of 0-1m, 1-2m, 2-3m, 3-4m, 4-5m and more than 5m correspond to different categories.

In this embodiment, the attribute data of the graph dataset is represented by a two-dimensional matrix, the dimension is 200×12, the first column is the index of the space node, and the last column is the category label corresponding to the groundwater level buried depth of the space node. The middle 10 columns are the eigenvectors of each spatial node. The edge connection data of the graph dataset is represented as a two-dimensional matrix with a dimension of 2315×2, which means that 200 spatial nodes have 2315 connected edges, and each row represents the indices of two connected nodes.

Further, in the present embodiment, the improvement of the GCN in the above-mentioned S4 is: the GCN network is improved, and the distance weight and attention mechanism are added, so that it can better utilize the surrounding neighbor features to learn the characteristics of the node itself; including: A weight matrix is added to the graph convolution operation of the first layer and the last layer, and the specific implementation is that the space node introduces a distance factor as an aggregation coefficient in the stage of aggregating neighbor node features. A graph attention convolution module is added after the first graph convolution layer. The input feature of this module is the feature output after the first layer graph convolution operation aggregates neighbor nodes. The output feature of this module is to automatically learn and update the node and its neighbor nodes. Aggregated features after weighting. After obtaining the feature output of the graph attention convolution module, it is fused with the output features of the first graph convolution layer, and the fused features are input into the last graph convolution layer.

Specifically, in this embodiment, the implementation process of improving GCN is as follows:

First of all, it should be noted that the proposal of GCN is to solve the problem that the convolution kernel cannot be shared in the convolution of irregular data. The traditional convolution adopts local perception area and shared weights, which can well extract the spatial features of the image. The graph structure does not have the translation invariance of the picture, so the traditional convolution method is not suitable for the graph structure. GCN is a first-order local approximation of spectral graph convolution. Neighborhood information is aggregated through convolutional layers, and multi-order neighborhood information transfer can be achieved by stacking several convolutional layers. Multiply to get a summary of the neighbor features of each vertex, and then multiply it by a trainable parameter matrix W. The adjacency matrix is normalized by the degree matrix to prevent some vertices with high degree and low degree from producing larger feature distributions. difference.

由节点邻接矩阵与单位矩阵I求和所得，

表示根据

按行求和所得的度矩阵，W^(l)表示可训练的参数矩阵，Dist为引入的距离权重矩阵，由于Dist对角线元素都为0，因此需要与单位矩阵I求和来保证在聚合邻居节点特征的过程中将中心节点自身的特征也考虑进去。Based on the above, the first improvement of this embodiment is to add edge connection weights in the graph convolution process. The weights are represented by the reciprocal of the geographic space distance. The function is that each node not only needs to consider the neighbor nodes when aggregating the neighbor node information The degree of , and the distance between the neighbor node and this node should also be considered. It can realize that the neighbor node with a farther distance has a smaller contribution to the node, and the neighbor node with a closer distance has a larger contribution to the node. The GCN calculation process after adding the distance weight is as follows, where H ^(l+1) represents the feature representation of the central node after one convolution,

It is obtained by summing the node adjacency matrix and the identity matrix I,

means according to

The degree matrix obtained by row-wise summation, W ^(l) represents the trainable parameter matrix, and Dist is the introduced distance weight matrix. Since the diagonal elements of Dist are all 0, it needs to be summed with the identity matrix I to ensure that in the aggregation In the process of neighbor node characteristics, the characteristics of the center node itself are also taken into account.

The second improvement made to GCN in this embodiment is to add a graph attention module. The calculation method of the attention factor in the attention module is shown in Figure 3. The network input is the graph structure data set constructed above, Graph represents the constructed graph structure data, X represents the adjacency matrix corresponding to the graph node, the solid point indicates that the corresponding node has an edge connection relationship, and the hollow point indicates that there is no edge connection relationship between nodes; After the first graph convolution layer, an attention module is added, and the output features of the attention module are fused with the output features of the graph convolution layer. This improvement enables spatial nodes to more accurately utilize the information of surrounding neighbor nodes and improve prediction. The overall structure of the improved GCN model network is shown in Figure 4.

Specifically, in this embodiment, after the spatial node features and their edge connection information are input into the first graph convolution layer, the graph node structure does not change, and each node generates a feature representation with a length of 32. In this process, the parameter matrix The dimension of W is 10×32, corresponding to the input feature dimension and the output feature dimension respectively. The features output by the first graph convolution layer are processed by the ReLU activation function, which can speed up model training, overcome the problem of gradient disappearance, and then use dropout processing to prevent model overfitting.

表示一个前馈神经网络，可通过训练更新参数，W表示参数矩阵，

表示节点的特征表示，

表示中心节点的邻居节点总数，特征相乘并拼接后通过LeakyReLU非线性化，经过Softmax归一化得到注意力系数。The node features after dropout processing in the first graph convolutional layer are input into the graph attention module. Since GCN cannot assign different weights according to the importance of neighbor nodes, although the inverse of the spatial distance is added in the first convolutional layer as The weights of different nodes, but the weight information cannot be learned and updated during the network training process, while the graph attention layer can learn and update the weight coefficients of nodes independently through training, which can better learn the dependencies between global features. The attention factor is calculated as follows, where

Represents a feedforward neural network, which can update parameters through training, W represents the parameter matrix,

represents the feature representation of the node,

Represents the total number of neighbor nodes of the central node. After the features are multiplied and spliced, they are nonlinearized by LeakyReLU and normalized by Softmax to obtain the attention coefficient.

After each node of the graph attention layer aggregates the features of the surrounding neighbor nodes through the attention factor, it also generates a node feature representation with a dimension of 32. In this embodiment, the features output by the graph attention module are combined with the first graph convolution layer. The output features are fused, and the fused feature dimension is n×64, where n represents the number of spatial nodes. This method can obtain enhanced representation of spatial node features. The fused enhanced features are then input into a graph convolution layer, the output is normalized by the log_softmax exponential function, the output is mapped to the range of 0-1, and then the loss in the model training process is calculated by NLLLoss, and backpropagation is used. Algorithms to correct network model parameters.

When training the groundwater level height prediction model, it is necessary to read the spatial node feature matrix and edge connection matrix into the memory, perform one-hot encoding on the node category, construct the adjacency matrix and degree matrix of the edge, and calculate the adjacency matrix and its transpose matrix. And, convert the directed graph into an undirected graph, and normalize the feature matrix and degree matrix. When dividing the training set, the validation set and the test set, the space nodes are randomly scrambled, divided according to the ratio of 3:1:2, the training set is input into the model for training, and the model is verified by the validation set in each training epoch. Effectiveness, the test set is used to test the prediction accuracy of the model after all training epochs, and the loss of the training set is calculated to update the model parameters through the back-propagation algorithm. The losses on the validation and test sets are not used for model optimization. Specifically, the training process of the above-mentioned groundwater level prediction model is as follows:

Repeat the above steps to train the improved GCN network until the loss is stable and no longer decreases, the model converges, and the optimal training parameters are obtained.

Further, in this embodiment, the implementation process of the above S5 is as follows:

Further, after realizing the prediction of the groundwater level at the spatial location, the method further includes:

Evaluate the validity of the prediction results of the groundwater level prediction model. Specifically, in this embodiment, the validity evaluation index of the groundwater level height prediction model is the accuracy rate, and the accuracy rate (Accuracy) refers to the proportion of the correctly predicted spatial node category to the total number of nodes in the test set. The calculation method of the accuracy rate is as follows:

In this example, two datasets with different surface water characteristics and groundwater level burial are constructed respectively in the wet season and the dry season, and the prediction model of the groundwater level height is evaluated according to the above evaluation indicators. The evaluation results are shown in Table 1. In order to explore the performance of the model, different training batches and learning rate sizes are used to evaluate the effectiveness of the model.

Table 1 Evaluation results of groundwater level height prediction model

In Table 1 above, Epoch is the number of iterations when using the training set to train the model, and different training iterations have an impact on the prediction effectiveness of the model. Learning rate indicates the magnitude of the influence on parameter changes when the error is back propagated during the training process, and the size of the learning rate also affects the prediction effect of the model. It can be seen from the above table that under the condition that the number of iterations is 1500 and the learning rate is 0.005, the prediction model of groundwater level has good performance in both wet and dry periods. The confusion matrix of prediction results of spatial nodes is shown in Figure 5.

To sum up, this embodiment provides a method for predicting groundwater level height by using surface water resources and geographic space attributes. By constructing a graph structure data set with spatial location nodes as the unit, adding distance weights between nodes, introducing attention mechanism and feature fusion mechanism, the method can obtain better feature expression of nodes. At the same time, the method of this embodiment can be used in some nodes When the groundwater level is unknown, the semi-supervised learning method is used to train the model, so as to realize the accurate prediction of the groundwater level in the spatial location, and solve the problems of inability to quantify and lack of accuracy in the existing surface water and groundwater evolution methods.

Second Embodiment

This embodiment provides a groundwater level height prediction system based on a multi-feature map network. The structure of the groundwater level height prediction system based on the multi-feature map network is shown in Figure 6, and includes the following modules:

Graph-structured dataset building blocks for:

Selecting a plurality of groundwater level monitoring points at different locations as spatial nodes for constructing a graph structure data set, acquiring characteristic information of each node, and constructing a characteristic vector of each node based on the characteristic information;

Weigh the correlation between the two nodes based on the feature vector, construct edge structure information for describing the connectivity of the nodes, and use the inverse of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

Using the feature information and edge structure information of the nodes as the sample features, and the depth interval category to which the groundwater level corresponding to the nodes belongs as the sample labels, a graph structure data set is constructed;

The groundwater height prediction model construction and training module is used to improve the graph convolutional network (Graph Convolutional Network, GCN) to obtain the groundwater level prediction model; groundwater level prediction model for training;

The groundwater height prediction module is used to construct the groundwater level prediction model trained by the groundwater height prediction model and the training module, so as to realize the prediction of the groundwater level of the spatial location;

The model evaluation module is used to evaluate the validity of the prediction results of the groundwater level prediction model.

The groundwater level height prediction system based on the multi-feature map network of this embodiment corresponds to the groundwater level height prediction method based on the multi-feature map network of the first embodiment; wherein, the groundwater level height prediction system based on the multi-feature map network The functions implemented by each functional module in the above correspond one by one to each process step in the above-mentioned method for predicting groundwater level height based on a multi-feature map network; therefore, the details are not repeated here.

Third Embodiment

This embodiment provides an electronic device, which includes a processor and a memory; wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the method of the first embodiment.

The electronic device may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPU) and one or more memories, wherein the memory stores at least one instruction, so The instructions are loaded by the processor and execute the above method.

Specifically, as shown in FIG. 7 , the electronic device may include a processor (CPU) 701 which may be loaded into a random access memory (RAM) 703 according to a program stored in a read only memory (ROM) 702 or from a storage part 708 to execute various appropriate actions and processes. A processor may include a general-purpose microprocessor (eg, a CPU), an instruction set processor and/or related chipset, and/or a special-purpose microprocessor (eg, an application specific integrated circuit (ASIC)), among others. The processor may also include onboard memory for caching purposes.

In addition, the device may also include an input/output (I/O) interface 705 , which is also connected to the bus 704 . and may also include one or more of the following components connected to the I/O interface 705: an input portion 706 including a keyboard, mouse, etc.; including components such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc. An output section 707 of the ; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, and the like. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I/O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 710 as needed so that a computer program read therefrom is installed into the storage section 708 as needed.

Fourth Embodiment

This embodiment provides a computer-readable storage medium, where at least one instruction is stored, and the instruction is loaded and executed by a processor to implement the method of the first embodiment. Wherein, the computer-readable storage medium may be ROM, random access memory, CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like. The instructions stored therein can be loaded by the processor in the terminal and execute the above method.

Furthermore, it should be noted that the present invention may be provided as a method, an apparatus or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, embedded processor or other programmable data processing terminal to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing terminal produce Means implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams. These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

It should also be noted that, herein, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or terminal device comprising a series of elements includes not only those elements, but also other elements not expressly listed or inherent to such process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

Finally, it should be noted that the above are the preferred embodiments of the present invention. It should be pointed out that although the preferred embodiments of the present invention have been described, for those skilled in the art, once the basic inventive concept of the present invention is known , without departing from the principles of the present invention, several improvements and modifications can also be made, and these improvements and modifications should also be regarded as the protection scope of the present invention. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims (10)

1. A groundwater level height prediction method based on a multi-feature map network is characterized by comprising the following steps:

selecting multiple underground water level monitoring points at different positions as space nodes for constructing a graph structure data set, acquiring characteristic information of each node, and respectively constructing a characteristic vector of each node based on the characteristic information;

based on the characteristic vector, weighing the correlation relationship between the two nodes, constructing edge structure information for describing the connectivity of the nodes, and taking the reciprocal of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

constructing a graph structure data set by taking the characteristic information and the edge structure information of the nodes as sample characteristics and taking the depth interval category to which the underground water level burial depth corresponding to the nodes belongs as a sample label;

improving a Graph Convolution Network (GCN) to obtain an underground water level prediction model; training the underground water level prediction model based on the graph structure data set;

and predicting the underground water level of the spatial position based on the trained underground water level prediction model.

2. The method for groundwater level height prediction based on a multi-feature map network as claimed in claim 1, wherein the feature information comprises surface water resource information and geospatial information;

the surface water resource information comprises irrigation quantity, water discharge quantity and total precipitation quantity in a preset period of the area where the node is located; the geospatial information comprises: the lithology of the aeration zone, the landform type, the permeability K1 partition, the name of the landform, the total solid solubility, the permeability coefficient KCP, and the distance between the node and the branch channel.

3. The method for predicting the height of the groundwater level based on the multi-feature map network according to claim 1, wherein the step of respectively constructing the feature vector of each node based on the feature information comprises the following steps:

vectorizing the characteristic information to obtain a characteristic vector of each node; the vectorization processing mode of the characteristic information is one-hot coding and normalization, wherein the one-hot coding and normalization processing are performed on unquantized information, and the normalization processing is performed on the quantized information directly.

4. The method for predicting the height of the groundwater level based on the multi-feature map network according to claim 1, wherein the step of constructing the edge structure information for describing the connectivity of the nodes by weighing the correlation relationship between two nodes based on the feature vectors comprises the following steps:

calculating cosine similarity, Pearson correlation coefficient and Euclidean distance among the feature vectors of all the nodes;

constructing a one-way connection relation graph of all nodes, reserving connection edges meeting preset conditions, and regarding nodes corresponding to the connection edges not meeting the preset conditions as non-connection relations; the preset conditions are that the cosine similarity between the characteristic vectors is larger than 0.7, the Pearson correlation coefficient is larger than 0.8, and the Euclidean distance is smaller than 1;

and counting the reserved connection relation after screening to obtain the side structure information for describing the node connectivity.

5. The method according to claim 1, wherein the depth interval categories include 0-1m, 1-2m, 2-3m, 3-4m, 4-5m, and 5m or more.

6. The method for predicting the height of the groundwater level based on the multi-feature map Network according to claim 1, wherein the improvement of a Graph volume Network (GCN) comprises:

adding a weight matrix in the convolution operation of the first layer and the last layer of the graph of the GCN network, and specifically realizing that: and the spatial node introduces edge connection weight as an aggregation coefficient in the aggregation neighbor node characteristic stage.

7. The method for predicting the height of the groundwater level based on the multi-feature map Network according to claim 6, wherein a Graph volume Network (GCN) is improved, and the method further comprises:

adding a graph attention convolution module after a first graph convolution layer of the GCN, wherein the input feature of the graph attention convolution module is the feature output after a neighbor node is aggregated by a first layer of graph convolution operation, and the output feature of the graph attention convolution module is the aggregated feature after the weights of the node and the neighbor node are automatically learned and updated; and after the feature output of the graph attention convolution module is obtained, the feature output is fused with the output feature of the first graph convolution layer, and the fused feature is input into the last graph convolution layer.

8. The multiple feature map network-based groundwater level height prediction method according to claim 1, wherein after the prediction of the groundwater level of the spatial location is achieved, the method further comprises:

and carrying out effectiveness evaluation on the prediction result of the underground water level prediction model.

9. The method for predicting the height of the underground water level based on the multi-feature map network as claimed in claim 8, wherein the evaluating the effectiveness of the prediction result of the underground water level prediction model comprises:

and comparing the types of the predicted result and the real result, and evaluating the effectiveness of the model through the prediction accuracy.

10. A groundwater level height prediction system based on a multi-feature map network is characterized by comprising:

a graph structure data set construction module to:

selecting multiple underground water level monitoring points at different positions as space nodes for constructing a graph structure data set, acquiring characteristic information of each node, and respectively constructing a characteristic vector of each node based on the characteristic information;

based on the characteristic vector, weighing the correlation relationship between the two nodes, constructing edge structure information for describing the connectivity of the nodes, and taking the reciprocal of the geographic distance between the two connected nodes as the edge connection weight of the nodes;

taking the characteristic information and the edge structure information of the nodes as sample characteristics, and taking the depth interval category to which the underground water level burial depth corresponding to the nodes belongs as a sample label to construct a graph structure data set;

the underground water height prediction model building and training module is used for improving a Graph Convolution Network (GCN) to obtain an underground water level prediction model; training the underground water level prediction model based on the graph structure data set constructed by the graph structure data set construction module;

and the underground water height prediction module is used for constructing an underground water level prediction model trained by the training module based on the underground water height prediction model and realizing the prediction of the underground water level of the spatial position.

Images