GraphSAGE-LSTM-based heavy fog nowcasting method and system

Abstract

The invention provides a GraphSAGE-LSTM-based heavy fog nowcasting method and system, and the method comprises the steps: collecting original meteorological data, detecting and eliminating abnormal data in the original meteorological data according to a preset time interval and threshold data, and obtaining suitable meteorological data; selecting at least two pieces of suitable heavy fog generation process information, collecting spatio-temporal information, processing according to the spatio-temporal information, the suitable meteorological data and the suitable heavy fog generation process information to obtain time sequence data, and constructing a data set according to the time sequence data; meteorological observation stations are used as graph nodes, preset logic and inter-node distance thresholds are utilized to construct an adjacency matrix, and an observation station graph structure is obtained; and constructing and utilizing an inductive graph neural network model, processing to obtain spatial distribution characteristics of surrounding meteorological elements, and processing to obtain a heavy fog nowcasting result. The technical problem that it is difficult to effectively extract spatial information between site data is solved.

Description

Technical Field

[0001] The present invention relates to the field of weather forecast data processing, and in particular to a heavy fog nowcasting method and system based on GraphSAGE-LSTM (GraphSample and Aggregate-Long Short-Term Memory). Background Art

[0002] The nowcast of fog can provide sufficient time for traffic management departments to respond to disasters and can also remind drivers to take precautions in advance. Many scholars have used the method of time series network to study the forecasting method of fog, but the time series network cannot well reflect the spatial variation characteristics of meteorological elements. The changes in meteorological elements at stations in the upstream and surrounding areas have an indicative effect on the nowcast of fog. The spatial distribution of station data belongs to a non-Euclidean structure, and traditional convolutional neural networks cannot effectively extract the spatial information between station data.

[0003] Traditional technologies have the following technical difficulties: (1) Construction of network model. Different from graph structures such as traffic flow prediction and recommendation systems, the graph structure is neither static nor dynamic due to the uncertainty of automatic weather station data, such as data anomalies, station failures, and the addition or cancellation of stations. It is necessary to select appropriate graph neural network structures and parameters.

[0004] (2) Real-time requirements: Nowcasting requires a high speed of model operation. It is necessary to consider how to generate prediction results within a limited time while ensuring the accuracy of the prediction.

[0005] The existing invention patent application document with publication number CN111784022A, "A short-term neighboring fog prediction method based on a combination of a Wrapper method and an SVM (Support Vector Machine) method", includes: the first step: obtaining the actual meteorological element attribute data at the current moment in the real-time system related to fog prediction; the second step: preprocessing the actual meteorological element attribute data; the third step: combining the Wrapper method and the SVM method to analyze the main characteristic attributes of the preprocessed data in the second step, and obtaining the main characteristic attribute data that can affect the fog prediction within different warning times; the fourth step: inputting the main characteristic attribute data in the third step into the SVM model with different warning times to obtain whether the data is foggy at different short-term neighboring moments. However, the model mentioned in the above-mentioned prior art does not take into account the changing characteristics of the meteorological elements of the surrounding stations, and only uses the time characteristics of the meteorological elements to predict future visibility changes. The spatial changes of meteorological elements are not fully considered.

[0006] The existing invention patent application document with publication number CN118483770A, "A wind speed prediction method for multiple wind farms with multiple features based on graph embedding and GIN-GRU (Graph Isomorphism Network-Gated Recurrent Unit)," adopts a deep learning method based on GraphSAGE for graph embedding. However, the existing technology combines the two models of GraphSAGE (Graph Sample and Aggregate) and GIN (Graph Isomorphism Network), which significantly increases the complexity of the model. At the same time, changes in the number of meteorological stations will cause changes in the graph structure. At the same time, the GIN model focuses on the overall spatial characteristics of the entire graph and cannot be applied to dynamic graph training.

[0007] In summary, the existing technology has the technical problem of being difficult to effectively extract spatial information between site data. Summary of the invention

[0008] The technical problem to be solved by the present invention is: how to solve the technical problem that it is difficult to effectively extract spatial information between site data in the prior art.

[0009] The present invention adopts the following technical solution to solve the above technical problem: A fog nowcasting method based on GraphSAGE-LSTM includes: S1. Collecting original meteorological data, detecting and removing abnormal data in the original meteorological data according to preset time intervals and threshold data, and obtaining applicable meteorological data; S2, selecting no less than 2 pieces of applicable fog occurrence process information, collecting spatiotemporal information, and processing the spatiotemporal information, applicable meteorological data, and applicable fog occurrence process information to obtain time series data, and constructing a data set accordingly; S3, taking the meteorological observation sites as graph nodes, using preset logic and distance thresholds between nodes, constructing an adjacency matrix to obtain the graph structure of the observation sites; S4. Construct and use an inductive graph neural network model to extract the spatial features of each graph node in each batch from the observation site graph structure, and use the time series neural network to extract the time features of the graph nodes in each batch. According to the spatial and time features, new node labels are obtained from the local neighbor information. The spatial distribution characteristics of the surrounding meteorological elements are processed using the time series data and the new node labels, and the fog nowcast results are obtained based on them.

[0010] The present invention constructs a graph structure of meteorological observation sites based on the k-nearest neighbor method to describe the spatial characteristics and dependencies between meteorological observation sites. On this basis, an inductive graph neural network model combining a graph neural network model and a long short-term memory neural network is used to perform visibility forecasting. Since the structure of the graph and the number of sites are neither static nor change all the time, the present invention selects an inductive graph neural network model to eliminate abnormal data in the original meteorological data, and can generate prediction results within a limited time while ensuring the accuracy of the prediction operation.

[0011] In a more specific technical solution, in S2, according to the preset data set selection rule, the applicable heavy fog occurrence process information is selected from the heavy fog single station data; The basic data set is formed based on the applicable fog occurrence process information and spatiotemporal information; According to the preset ratio, the basic data set is divided into training set, test set and validation set.

[0012] In a more specific technical solution, in S3, the k-nearest neighbor algorithm KNN (K-Nearest Neighbor) is used to select no less than 2 meteorological observation sites according to the distance threshold between nodes, and then construct an adjacency matrix.

[0013] In a more specific technical solution, in S4, input data of the inductive graph neural network model is set, wherein the input data includes: node features and an adjacency matrix; Get the visibility of the input data and adjust the visibility V Logarithmic processing is performed to obtain visibility parameters :

[0014] Perform Max-Min normalization on all elements in the input data; In the adjacency matrix, an undirected unweighted matrix is ​​used to define the neighbor relationship of graph nodes; An encoder and a GraphSAGE module of the inductive graph neural network model are set. In the GraphSAGE module, neighbor node features are sampled according to neighbor relationships to perform node representation update operations to obtain spatial distribution features of surrounding meteorological elements. The GraphSAGE module uses two-order neighbor sampling to capture the spatial relationship of the applicable scope to obtain the spatial distribution features of surrounding meteorological elements. The neighbor node features are aggregated through the mean aggregation operation. In this case, the inductive graph neural network model is used to combine the current time step with the LSTM gating mechanism in each time step. t Given input features x t, and the aggregation information of neighbor nodes, update the hidden state and cell state of the graph node; Set up the decoder of the inductive graph neural network model and use the decoder to convert the hidden state h t Convert to output; Using the following logic, set the loss function of the inductive graph neural network model:

[0015] in, F(y) is the predicted cumulative distribution function; is the real observation value, obs represents the observation operation; y is the predicted value; 1 is the indicator function, when the predicted value y The actual observed value Satisfy y ≥ y obs When y 1, otherwise the predicted value y is 0.

[0016] In the process of constructing the inductive graph neural network model, the present invention selects a graph neural network model that can handle new nodes and sites. The inductive graph neural network model used in the present invention can infer the label of the new node from the local neighbor information without retraining the entire model.

[0017] In a more specific technical solution, the following logic is used to set the input data X:

[0018] in, B is the batch size, N is the number of graph nodes, T is the number of time steps, F is the feature dimension at each time step, Represents the batch size B , graph node set N , the number of time steps T , the feature dimension at each time step F The corresponding number field.

[0019] In a more specific technical solution, for a given input feature x , using the following logic, the input data Max-Min is normalized to obtain the normalized features :

[0020] in, represents the maximum given input feature, Represents the smallest given input feature.

[0021] In a more specific technical solution, the neighbor relationship is defined using the following logic: : .

[0022] In a more specific technical solution, the following logic is used to perform node representation update operations, and the update formula for each layer is:

[0023] in, yes Relu Activation function; W k It is k The learnable weight matrix of the layer W ; N (i) Indicates i The set of neighbor nodes of a node, Indicates k Layer i The hidden state corresponding to the node h , Indicates k-1 Layer i The hidden state corresponding to the node h , Indicates k-1 Layer j The hidden state corresponding to the node h , MEAN Represents an update operation.

[0024] The inductive graph neural network model adopted in the present invention combines the advantages of GraphSAGE and LSTM models. It can not only capture the time evolution characteristics of nodes, but also aggregate neighbor information through GraphSAGE to capture the spatial characteristics of the graph structure.

[0025] The present invention directly uses GraphSAGE for sampling and obtains the local spatial features of the graph to input into the LSTM model, paying more attention to the changes in local spatial features, which can reduce the impact of changes in the overall graph structure on the model accuracy. At the same time, in the process of forecasting local fog, the present invention pays attention to the changes in meteorological elements of adjacent sites, which can effectively extract local information and improve the accuracy of fog forecasting.

[0026] The GraphSAGE model used in the present invention updates the representation of nodes by sampling the features of neighboring nodes. The module uses two-order neighbor sampling, that is, capturing a wider range of spatial relationships through neighbors and neighbors of neighbors.

[0027] In a more specific technical solution, the following logic is used to set the forget gate :

[0028] Using the following logic, set the input gate :

[0029] Use the following logic to set the candidate cell state :

[0030] Update the cell state using the following logic :

[0031] Using the following logic, set the output gate :

[0032] Use the following logic to update the hidden state :

[0033] in; is the hidden state at the previous moment; W f is the weight matrix of the forget gate; W input is the weight matrix of the input gate; W C is the weight matrix of the cell state; W o is the weight matrix of the output gate; b f is the bias vector of the forget gate; b input is the bias vector of the input gate; b C is the bias vector of the cell state; b o is the bias vector of the output gate; σ is Relu Activation function.

[0034] In a more specific technical solution, a fog nowcasting system based on GraphSAGE-LSTM includes: The data collection and quality control module is used to collect raw meteorological data, detect and remove abnormal data in the raw meteorological data according to preset time intervals and threshold data, and obtain applicable meteorological data; A data set construction module is used to select no less than 2 pieces of applicable fog occurrence process information, collect spatiotemporal information, process the spatiotemporal information, applicable meteorological data and applicable fog occurrence process information to obtain time series data, and construct a data set based on the data set construction module, which is connected to the data collection and quality control module; The observation station graph structure construction module is used to use the meteorological observation stations as graph nodes, use the preset logic and the distance threshold between nodes to construct the adjacency matrix, and obtain the observation station graph structure. The observation station graph structure construction module is connected with the data set construction module; The module for acquiring the distribution characteristics of surrounding meteorological elements is used to construct and use the inductive graph neural network model to extract the spatial characteristics of each graph node in each batch from the observation site graph structure, and use the time series neural network to extract the time characteristics of the graph nodes in each batch. According to the spatial characteristics and time characteristics, new node labels are obtained from the local neighbor information. The spatial distribution characteristics of surrounding meteorological elements are processed using time series data and new node labels, and the fog nowcast results are obtained based on this processing. The module for acquiring the distribution characteristics of surrounding meteorological elements is connected to the observation station graph structure construction module.

[0035] Compared with the prior art, the present invention has the following advantages: The present invention constructs a graph structure of meteorological observation sites based on the k-nearest neighbor method to describe the spatial characteristics and dependencies between meteorological observation sites. On this basis, an inductive graph neural network model combining a graph neural network model and a long short-term memory neural network is used to perform visibility forecasting. Since the structure of the graph and the number of sites are neither static nor change all the time, the present invention selects an inductive graph neural network model to eliminate abnormal data in the original meteorological data, and can generate prediction results within a limited time while ensuring the accuracy of the prediction operation.

[0036] In the process of constructing the inductive graph neural network model, the present invention selects a graph neural network model that can handle new nodes and sites. The inductive graph neural network model used in the present invention can infer the label of the new node from the local neighbor information without retraining the entire model.

[0037] The inductive graph neural network model adopted in the present invention combines the advantages of GraphSAGE and LSTM models. It can not only capture the time evolution characteristics of nodes, but also aggregate neighbor information through GraphSAGE to capture the spatial characteristics of the graph structure.

[0038] The present invention directly uses GraphSAGE for sampling and obtains the local spatial features of the graph to input into the LSTM model, paying more attention to the changes in local spatial features, which can reduce the impact of changes in the overall graph structure on the model accuracy. At the same time, in the process of forecasting local fog, the present invention pays attention to the changes in meteorological elements of adjacent sites, which can effectively extract local information and improve the accuracy of fog forecasting.

[0039] The GraphSAGE model used in the present invention updates the representation of nodes by sampling the features of neighboring nodes. The module uses two-order neighbor sampling, that is, capturing a wider range of spatial relationships through neighbors and neighbors of neighbors.

[0040] The present invention solves the technical problem in the prior art that it is difficult to effectively extract spatial information between site data. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of basic steps of a heavy fog nowcasting method based on GraphSAGE-LSTM according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the change of training loss in Example 1 of the present invention; Figure 3a This is a schematic diagram of a heavy fog forecast for the I9837 station according to Embodiment 2 of the present invention; Figure 3b This is a schematic diagram of a heavy fog forecast for the I9699 site according to Embodiment 2 of the present invention; Figure 3c This is a schematic diagram of a heavy fog forecast for the I9893 site according to Embodiment 2 of the present invention; Figure 3d This is a schematic diagram of the forecast of heavy fog for four stations I0026 in Example 2 of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Example 1 like Figure 1 As shown, the present invention provides a GraphSAGE-LSTM-based fog nowcasting method, which includes the following basic steps: S1. Collecting original meteorological data, detecting and removing abnormal data in the original meteorological data according to preset time intervals and threshold data, and obtaining applicable meteorological data; In this embodiment, the original data is processed into, for example, 10-minute data, and the original data is resampled at 10-minute intervals. At the same time, the visibility, temperature, relative humidity, wind speed, rainfall and meteorological elements of surrounding stations are statistically analyzed to detect and eliminate abnormal data. Specific data quality control rules include but are not limited to: The number of data is 10, and the wind direction, wind speed, precipitation, temperature, relative humidity, and visibility data cannot be missing; When the median visibility is less than 500 meters, the relative humidity is not less than 80%, and the wind speed cannot be greater than, for example: 5.4m / s; The median visibility cannot be 0, and the standard deviation cannot be 0.

[0044] S2, selecting no less than 2 pieces of applicable fog occurrence process information, collecting spatiotemporal information, and processing the spatiotemporal information, applicable meteorological data, and applicable fog occurrence process information to obtain time series data, and constructing a data set accordingly; In this embodiment, according to the preset data set selection rules, the applicable fog occurrence process information is selected from the fog single station data; The basic data set is formed based on the applicable fog occurrence process information and spatiotemporal information; According to the preset ratio, the basic data set is divided into training set, test set and validation set.

[0045] In this embodiment, the visibility level is predicted every 10 minutes within 0-2 hours, and the time series data is extracted for model training. Using the 10-minute data of the past 2 hours to predict the visibility changes in the next 2 hours, it is necessary to extract 24 consecutive time data as an example. The data set selection rules include but are not limited to: 24 consecutive time periods, a total of 4 hours of time series data, all passed the quality control in the above step S1; The visibility at any meteorological observation station is less than 500 meters for no less than 3 times; At least once in a while, the visibility at more than 10 meteorological observation stations is less than 500 meters; The interval between two adjacent cases shall not be less than 1 hour.

[0046] In this embodiment, 4626 fog processes, a total of 186 single-station fog data, are selected, and the original observation data and spatiotemporal information of the meteorological observation stations are selected to form a basic data set. The training set, test set, and validation set are obtained by dividing according to the principle of 8:1:1, for example. In this embodiment, the original observation data include but are not limited to: temperature, relative humidity, wind direction and speed, air pressure, precipitation, and visibility; the spatiotemporal information includes but is not limited to: altitude, longitude and latitude, and solar altitude angle.

[0047] S3, using meteorological observation sites as graph nodes, using preset logic and node distance thresholds, constructing an adjacency matrix to obtain the observation site graph structure; specifically, using the k-nearest neighbor algorithm KNN, according to the node distance threshold, selecting no less than 2 meteorological observation sites, and constructing the adjacency matrix accordingly; In this embodiment, the node is each meteorological observation site, and the k-nearest neighbor algorithm is used to select the five nearest meteorological sites as neighbor nodes to construct an adjacency matrix, and at the same time, the distance between neighbor nodes is required to be no more than 20 kilometers.

[0048] S4. Construct and use an inductive graph neural network model to extract the spatial features of each graph node in each batch from the observation site graph structure, and use a time series neural network to extract the time features of the graph nodes in each batch. According to the spatial features and time features, new node labels are processed from local neighbor information. The spatial distribution characteristics of the surrounding meteorological elements are processed using time series data and new node labels, and the fog nowcast results are obtained based on the processing; In this embodiment, due to data quality, instrument failure, and network transmission, the number of meteorological observation sites in actual business may be inconsistent with the number of samples when building the model. Therefore, in the model construction process, a graph neural network model that can handle new nodes is selected. The inductive graph neural network model can infer the label of the new node from the local neighbor information without retraining the entire model. Using the time series data of meteorological elements, and further considering the spatial distribution characteristics of the surrounding meteorological elements, an inductive graph neural network model is used to extract the spatial characteristics of each meteorological site data in each batch, and then the time series neural network is used to extract the time characteristics of the meteorological site in each batch. A visibility forecast of 10 minutes within 0-2h is achieved.

[0049] In this embodiment, the input data of the graph neural network model is set, including but not limited to: node features, adjacency matrix; In this embodiment, the node feature corresponds to each node i At 12 time steps t There is F dimensional features. The input shape of time series data is:

[0050] in, B is the batch size; N is the number of nodes; T is the number of time steps, which can be, for example, 12; F is the feature dimension of each time step, which can be taken as 10, for example. The surrounding meteorological elements include temperature, relative humidity, wind direction and speed, air pressure, precipitation, visibility, altitude, longitude, latitude, and solar altitude. Before inputting into the inductive graph neural network model, all surrounding meteorological elements are normalized by Max-Min. For a given input feature x , the formula for Max-Min normalization is:

[0051] In this embodiment, due to visibility V The maximum can be set to 30km, for example, and the minimum can be set to 0km, for example. The data span is large, and the visibility is low below 1km. V Logarithmic processing to reduce visibility V The above normalization process is performed on this basis:

[0052] In this embodiment, the adjacency matrix uses an undirected and weightless matrix to define the neighbor relationship of the nodes:

[0053] In this embodiment, an encoder of an inductive graph neural network model is set, and the node representation is updated by sampling neighbor node features using the GraphSAGE model. The aforementioned GraphSAGE model uses two-order neighbor sampling, that is, capturing a wider range of spatial relationships through neighbors and neighbors of neighbors. For each node, first sample the first-order neighbors, then sample the neighbors of these neighbors to obtain the second-order neighbors. N(N (i) ) . Use the mean aggregation operation to aggregate the features of neighboring nodes. For node i and its neighbor set N (i) , perform node representation update operation, the update formula for each layer is:

[0054] In this embodiment, the inductive graph neural network model combines the capabilities of the GraphSAGE model and the LSTM model, which can not only capture the time evolution characteristics of the node, but also capture the spatial characteristics of the graph structure by aggregating the characteristics of neighboring nodes through GraphSAGE. In each time step, the gating mechanism of the LSTM model combines the input features of the current time step and the aggregated information of neighboring nodes to update the hidden state and cell state of the node. Specifically: Forget Gate The settings are as follows:

[0055] Input Gate The settings are as follows:

[0056] Candidate cell states The settings are as follows:

[0057] Cell state The update method is as follows:

[0058] Output Gate The settings are as follows:

[0059] Hidden State The update method is as follows:

[0060] In this embodiment, the number of layers of the GraphSAGE-LSTM inductive graph neural network model can be set to, for example, 2 layers, and the number of cells can be set to, for example, 50.

[0061] In this embodiment, a decoder of the inductive graph neural network model is set. In this embodiment, the decoder includes: The fully connected layer is used to transform the hidden state h t Converted to output, the number of fully connected layers can be set to, for example, 1 layer, and the number of convolution kernels can be set to, for example, 100.

[0062] In this embodiment, the loss function of the inductive graph neural network model is set. For example, CRPS (Continuous Ranked Probability Score) can be used as the loss function of the aforementioned inductive graph neural network model, and the loss function is used to measure the distance between the probability distribution prediction and the true value. The smaller the CRPS, the closer the predicted distribution is to the true value. For details, see the following formula:

[0063] In this embodiment, the output data of the inductive graph neural network model is set; in this embodiment, the output is, for example: 12 time periods, 10 minutes each, of the visibility forecast of the weather station for the next 2 hours.

[0064] like Figure 2 As shown, in this embodiment, the inductive graph neural network model is trained and tested; the above-mentioned inductive graph neural network model is used to train the data set, and the loss curve is shown in Figure 2 , after 100 rounds of training, the loss on the validation set tends to be stable.

[0065] Example 2 In this embodiment, after the model training is completed, the random forest model, SVM model, CNN model, and LSTM model are used to compare and analyze the fog forecast results on the test set. The visibility is divided into 5 categories according to [0m, 50m], [50m, 100m], [100m, 200m], and [200m, 500m]. Visibility greater than 500m is divided into 5 categories. The three indicators of F1-Score, precision rate, and TS score are used to evaluate the random forest model, SVM (Support Vector Machine) model, CNN (Convolutional Neural Networks) model, and LSTM (Long Short-Term Memory) model. It is found that the GraphSAGE-LSTM inductive graph neural network model used in the present invention is higher than the random forest model, SVM model, CNN model, and LSTM model in all three indicators.

[0066]

[0067] Table 1 Scores of random forest, SVM, CNN, LSTM and this method on the test set like Figure 3a , Figure 3b , Figure 3c , Figure 3dAs shown, in this embodiment, case tests are carried out and the GraphSAGE-LSTM inductive graph neural network model has a strong predictive ability for visibility fluctuations during heavy fog, the rapid decrease in visibility in the initial stage of fog formation, the stable maintenance of visibility during heavy fog, and the rapid increase in visibility when the fog dissipates. There are certain differences between the visibility forecast and the actual data at individual times, but the overall change trend is basically consistent with the actual situation, and the difference is small, which can well predict the changes in visibility in the next 2 hours.

[0068] In summary, the present invention constructs a graph structure of meteorological observation sites based on the k-nearest neighbor method to describe the spatial characteristics and dependencies between meteorological observation sites. On this basis, an inductive graph neural network model combining a graph neural network model and a long short-term memory neural network is used to perform visibility forecasting. Since the structure of the graph and the number of sites are neither static nor change all the time, the present invention selects an inductive graph neural network model to eliminate abnormal data in the original meteorological data, and can generate prediction results within a limited time while ensuring the accuracy of the prediction operation.

[0069] In the process of constructing the inductive graph neural network model, the present invention selects a graph neural network model that can handle new nodes and sites. The inductive graph neural network model used in the present invention can infer the label of the new node from the local neighbor information without retraining the entire model.

[0070] The inductive graph neural network model adopted in the present invention combines the advantages of GraphSAGE and LSTM models. It can not only capture the time evolution characteristics of nodes, but also aggregate neighbor information through GraphSAGE to capture the spatial characteristics of the graph structure.

[0071] The present invention directly uses GraphSAGE for sampling and obtains the local spatial features of the graph to input into the LSTM model, paying more attention to the changes in local spatial features, which can reduce the impact of changes in the overall graph structure on the model accuracy. At the same time, in the process of forecasting local fog, the present invention pays attention to the changes in meteorological elements of adjacent sites, which can effectively extract local information and improve the accuracy of fog forecasting.

[0072] The GraphSAGE model used in the present invention updates the representation of nodes by sampling the features of neighboring nodes. The module uses two-order neighbor sampling, that is, capturing a wider range of spatial relationships through neighbors and neighbors of neighbors.

[0073] The present invention solves the technical problem in the prior art that it is difficult to effectively extract spatial information between site data.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fog nowcasting method based on GraphSAGE-LSTM, characterized in that: The method comprises: S1. Collecting original meteorological data, detecting and removing abnormal data in the original meteorological data according to preset time intervals and threshold data, and obtaining applicable meteorological data; S2, selecting no less than 2 pieces of applicable fog occurrence process information, collecting spatiotemporal information, and processing the spatiotemporal information, the applicable meteorological data, and the applicable fog occurrence process information to obtain time series data, and constructing a data set accordingly; S3, taking the meteorological observation sites as graph nodes, using preset logic and distance thresholds between nodes, constructing an adjacency matrix to obtain the graph structure of the observation sites; S4. Construct and use an inductive graph neural network model to extract the spatial features of each graph node in each batch from the observation site graph structure, and use a time series neural network to extract the time features of the graph nodes in each batch. According to the spatial features and the time features, new node labels are obtained from local neighbor information. The spatial distribution features of the surrounding meteorological elements are obtained by using the time series data and the new node labels, and the heavy fog nowcasting results are obtained based on the processing.

2. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 1 is characterized in that: In S2, according to the preset data set selection rule, the applicable heavy fog occurrence process information is selected from the heavy fog single station data; forming a basic data set according to the applicable fog occurrence process information and the spatiotemporal information; The basic data set is divided according to a preset ratio to obtain a training set, a test set, and a validation set.

3. The heavy fog nowcasting method based on GraphSAGE-LSTM according to claim 1 is characterized in that: In S3, the k-nearest neighbor algorithm KNN is used to select no less than two meteorological observation sites according to the node distance threshold, and the adjacency matrix is ​​constructed accordingly.

4. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 1, characterized in that: In S4, input data of the inductive graph neural network model is set, wherein the input data includes: node features and the adjacency matrix; Obtain the visibility in the input data, and V Logarithmic processing is performed to obtain visibility parameters : Performing Max-Min normalization processing on all elements in the input data; In the adjacency matrix, using an undirected unweighted matrix, the neighbor relationship of the graph nodes is defined; An encoder and a GraphSAGE module of the inductive graph neural network model are set, and in the GraphSAGE module, neighbor node features are sampled according to the neighbor relationship to perform a node representation update operation to obtain the spatial distribution characteristics of the surrounding meteorological elements, wherein the GraphSAGE module uses two-order neighbor sampling to capture the applicable range spatial relationship to obtain the spatial distribution characteristics of the surrounding meteorological elements; The neighbor node features are aggregated by the mean aggregation operation, wherein the inductive graph neural network model is used to combine the current time step with the LSTM gating mechanism in each time step. t Given input features x t , and the aggregation information of neighbor nodes, updating the hidden state and cell state of the graph node; Set a decoder of the inductive graph neural network model, and use the decoder to convert the hidden state h t Convert to output; The loss function of the inductive graph neural network model is set using the following logic: in, F(y) is the predicted cumulative distribution function; is the real observation value, obs represents the observation operation; y is the predicted value; 1 is the indicator function, when the predicted value y The true observed value Satisfy y ≥ y obs When y is 1, otherwise the predicted value y is 0.

5. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 4 is characterized in that: The input data X is set using the following logic: In the formula, B is the batch size, N is the number of nodes in the graph, T is the number of time steps, F is the feature dimension at each said time step, Represents the batch size B , the graph node set N , the number of time steps T , the feature dimension of each time step F The corresponding number field.

6. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 4 is characterized in that: For the given input features x , using the following logic, the input data is normalized to Max-Min to obtain the normalized feature : In the formula, represents the maximum of the given input features, represents the minimum of the given input features.

7. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 4 is characterized in that: The neighbor relationship is defined using the following logic: : 。 8. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 4 is characterized in that: The node representation update operation is performed using the following logic, and the update formula for each layer is: in, yes Relu Activation function; W k It is k The learnable weight matrix of the layer W ; N (i) Indicates i The set of neighbor nodes of a node, Indicates the k Layer i The hidden state corresponding to the node h , Indicates k-1 Layer i The hidden state corresponding to the node h , Indicates k-1 Layer j The hidden state corresponding to the node h , MEAN Represents an update operation.

9. The method for heavy fog nowcasting based on GraphSAGE-LSTM according to claim 4, characterized in that: Use the following logic to set the forget gate : Using the following logic, set the input gate : Use the following logic to set the candidate cell state : The cell state is updated using the following logic : Using the following logic, set the output gate : The hidden state is updated using the following logic : Where: is the hidden state at the previous moment; W f is the weight matrix of the forget gate; W input is the weight matrix of the input gate; W C is the weight matrix of the cell state; W o is the weight matrix of the output gate; b f is the bias vector of the forget gate; b input is the bias vector of the input gate; b C is the bias vector of the cell state; b o is the bias vector of the output gate; σ is Relu Activation function.

10. A heavy fog nowcasting system based on GraphSAGE-LSTM, characterized in that: The system comprises: The data collection and quality control module is used to collect raw meteorological data, detect and remove abnormal data in the raw meteorological data according to preset time intervals and threshold data, and obtain applicable meteorological data; A data set construction module is used to select no less than 2 pieces of applicable fog occurrence process information, collect spatiotemporal information, and process the spatiotemporal information, the applicable meteorological data, and the applicable fog occurrence process information to obtain time series data, thereby constructing a data set. The data set construction module is connected to the data acquisition and quality control module. An observation station graph structure construction module is used to use meteorological observation stations as graph nodes, use preset logic and distance thresholds between nodes to construct an adjacency matrix, and obtain an observation station graph structure. The observation station graph structure construction module is connected to the data set construction module; The module for acquiring the distribution characteristics of surrounding meteorological elements is used to construct and use an inductive graph neural network model to extract the spatial characteristics of each graph node in each batch from the observation site graph structure, and use a time series neural network to extract the time characteristics of the graph nodes in each batch. According to the spatial characteristics and the time characteristics, new node labels are obtained from local neighbor information, and the spatial distribution characteristics of surrounding meteorological elements are obtained by using the time series data and the new node labels. The heavy fog nowcasting result is obtained based on the processing. The module for acquiring the distribution characteristics of surrounding meteorological elements is connected to the observation station graph structure construction module.

Images