Sunspot area prediction method based on space-time diagram neural network
Through the method based on the spatiotemporal graph neural network, the space-time graph of sunspots is constructed and trained, which solves the problem of poor accuracy in the prediction of sunspot area in traditional prediction methods, and achieves higher accuracy and adaptability.
Patent Information
- Application Number
- CN202411059659.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-12-13
AI Technical Summary
The traditional sunspot area prediction method has poor prediction accuracy due to complex solar activity systems and multiple influencing factors, and there is great uncertainty.
Using a method based on a spatiotemporal graph neural network, we obtain solar observation data from an astronomical observation station, clean, process and integrate data, build a spatiotemporal graph, and use graph neural network model training to learn the spatiotemporal relationship of sunspots for prediction.
Effectively dealing with complex modes and nonlinear relationships, taking into account spatiotemporal information, improving the accuracy and reliability of predictions, reducing the impact of human factors on the prediction results, being highly adaptable, and being able to better capture the dynamic characteristics of solar activities.
Smart Images
Figure CN119128418A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of solar observation technology, and in particular to a sunspot area prediction method based on a spatiotemporal graph neural network. Background Art
[0002] The impact of the space environment on space activities has received increasing attention from technical departments. Whether it is spacecraft design, launch and recovery, or flight plan preparation, orbital flight attitude control operations, etc., the impact of the space environment needs to be considered. Solar activity is the source of disturbance in the space environment and the main cause of catastrophic events in the space environment. The level of solar activity directly affects the state of the near-Earth space environment (atmospheric environment, ionospheric environment, etc.). When solar activity changes dramatically, the atmospheric density can change by several times, seriously affecting normal orbit maintenance.
[0003] Sunspots, also known as solar spots, are temporary phenomena on the solar photosphere. They appear as darker spots than the surrounding areas under visible light. The formation and disappearance of sunspots can take anywhere from a few days to a few weeks. The darkest part in the center of a sunspot is called the umbra, which is the area with the strongest magnetic field. The less dark, striped area around the umbra is called the penumbra. Sunspots rotate with the surface of the sun and complete a rotation in about 27 days. Long-term observations have found that when there are many sunspots, other solar activity phenomena will also be more frequent. There will always be spots of light in the photosphere near sunspots, and there will always be spectral spots in the chromosphere above the sunspots, and there will often be prominences near them. Most solar eruption activity phenomena also occur in the atmosphere above sunspots. From the lower to the upper layers of the solar atmosphere, an activity center is formed with sunspots as the core, namely the solar active zone. Sunspots are both the core of the active zone and the most obvious sign of the active zone.
[0004] People usually use the relative number of sunspots (i.e., Wolf number) or the area of sunspots to study its changing patterns. Among them, the number of sunspots records the area of sunspots that appear on the visible solar disk every day (monthly average). In a sense, it records the daily magnetic flux of solar magnetic activity, which represents the power of the solar generator to generate sunspots (Li Kejun et al., 2004), with the unit of one millionth of the area of the solar hemisphere. Therefore, compared with the number of sunspots, the area of sunspots has more physical significance and is an important physical quantity that describes many solar activities.
[0005] A common method for predicting the traditional sunspot area is to use the periodic characteristics of the solar activity cycle for prediction. This method assumes that the appearance and disappearance of sunspots are periodic, and the periodic changes can be predicted. By analyzing the relationship between the past solar activity cycle and the sunspot area, a mathematical model is established to predict the future sunspot area. Another method is based on the observation of the solar magnetic field. Sunspots are magnetically active areas on the surface of the sun, so the observation of the solar magnetic field can provide clues for predicting sunspots. Scientists use data from solar observation satellites, telescopes, and ground observatories to monitor changes in the solar magnetic field, and combine these observations with known solar activity cycles to predict future sunspot areas. However, solar activity is a complex system, which is affected by many factors such as the physical processes inside the sun and the air flow movement on the surface of the sun, resulting in poor prediction accuracy and large uncertainty. To this end, we propose a sunspot area prediction method based on spatiotemporal graph neural network. Summary of the invention
[0006] The purpose of the present invention is to provide a sunspot area prediction method based on spatiotemporal graph neural network to solve the problems raised in the above background technology.
[0007] To achieve the above purpose, a sunspot area prediction method based on spatiotemporal graph neural network includes the following steps:
[0008] S1. Data collection: Obtain solar observation data from astronomical observatories, including the area, location and time information of sunspots;
[0009] S2, data processing: cleaning and preprocessing the sample data obtained in step S1, and then integrating the processed data to mine the associated solar state information to form multiple related data packets;
[0010] S3, constructing a space-time map: constructing a space-time map of sunspots based on observation data;
[0011] S4. Space-time graph neural network model training: Use the space-time graph neural network model to train the constructed space-time graph. The model will learn the space-time relationship of sunspots and model it with the sunspot area.
[0012] S5. Predict sunspot area: Use the trained space-time graph neural network model to predict the future sunspot space-time graph.
[0013] Preferably, the data processing in step S2 includes:
[0014] Data cleaning: the process of checking and processing errors, missing values, outliers and duplicate values in the acquired solar observation data. Data cleaning techniques include removing duplicate values, filling missing values, processing outliers and screening useful data.
[0015] Data conversion: convert solar observation data from raw formats to formats more suitable for analysis. Data conversion techniques include data encoding, format conversion, data normalization, and aggregation.
[0016] Data integration: Combining solar observation data from different sources and formats into a unified data set. Data integration techniques include data merging, joining and correlation.
[0017] Feature engineering: Extract, construct and select appropriate features from raw data based on the domain knowledge of solar observation data and problem requirements to improve the performance and effect of the model. Feature engineering techniques include feature extraction, feature conversion and feature selection;
[0018] Data reduction: When processing large-scale solar observation data, it is necessary to reduce the size and complexity of the data and improve computational efficiency. Data reduction techniques include data sampling, downsampling, and dimensionality reduction.
[0019] Data standardization: Convert solar observation data to the same scale and range to eliminate differences between data. Data standardization techniques include normalization and standardization.
[0020] Data Visualization: Use charts and graphic visualization tools to display and present solar observation data to help understand the distribution, relationships and trends of the data. Data visualization techniques include bar charts, line charts, scatter plots and heat maps.
[0021] Preferably, constructing the space-time graph in step S3 includes:
[0022] S31. Define nodes and edges: each sunspot is regarded as a node in the graph. The characteristics of the node are the area and position of the sunspot. According to the time relationship between sunspots, define edges to connect the nodes. The connection of the edges is determined by calculating the time interval between sunspots or other correlation indicators.
[0023] S32. Construction of space-time graph: Based on the defined nodes and edges, a space-time graph structure of sunspots is constructed. A graph data structure or a network data structure is used to represent the space-time graph. Each node represents the characteristics of a sunspot, and the edges represent the temporal relationship between sunspots.
[0024] S33, feature extraction and encoding: for each node and edge, feature extraction and encoding are performed;
[0025] S34. Representation and storage of space-time graph: The constructed space-time graph is represented as an adjacency matrix and stored in an appropriate data structure.
[0026] Preferably, the adjacency matrix is a method for representing the connection relationship between nodes in a graph data structure, and the method for obtaining the adjacency matrix includes an adjacency list representation method (Adjacency List) and an adjacency matrix representation method (Adjacency Matrix)
[0027] Adjacency List: In adjacency list representation, each node has a list of nodes adjacent to it. The steps to obtain the adjacency matrix include:
[0028] ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0;
[0029] ②Traverse each node in the graph, and for each node, find a list of nodes adjacent to it;
[0030] ③ Set the elements of the corresponding rows and columns in the matrix to 1, indicating that there is a connection relationship between the two nodes.
[0031] Adjacency Matrix: In the adjacency matrix representation, a two-dimensional matrix is used to represent the connection relationship between nodes in the graph. The steps to obtain the adjacency matrix include:
[0032] ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0;
[0033] ② Traverse each edge in the graph, and for each edge, set the corresponding element in the matrix to 1, indicating that there is a connection relationship between the two nodes;
[0034] ③If the graph is a weighted graph, the elements in the matrix can be set to the weight values of the edges, indicating the strength of the connection between the nodes.
[0035] Preferably, in step S4, a graph convolutional neural network is used to train the constructed space-time graph, and the model learns the space-time relationship of sunspots and models it with the sunspot area.
[0036] Preferably, the spatiotemporal graph neural network model training process in step S4 includes:
[0037] S41, data preparation: extract and prepare training data, including the spatiotemporal map data of sunspots and the corresponding sunspot areas;
[0038] S42. Model design: Design the structure of the spatiotemporal graph neural network model, including the model's hierarchy and connection methods;
[0039] S43, model initialization: initializing the weight parameters of the spatiotemporal graph neural network model;
[0040] S44, forward propagation: forward propagation of training data through the spatiotemporal graph neural network model to generate prediction results;
[0041] S45. Loss function definition: Define a loss function to measure the difference between the predicted result and the true label. The loss function includes mean square error (MSE) and cross entropy loss function.
[0042] S46, Back Propagation: Use the back propagation algorithm to calculate the gradient of the model parameters, and update the model parameters according to the gradient to minimize the loss function. The optimization algorithms include Stochastic Gradient Descent (SGD) and Adam.
[0043] S47, Iterative training: Through multiple iterations, repeat the forward propagation and back propagation steps, and use the training data to continuously optimize the model parameters so that the model can better fit the data.
[0044] Compared with the prior art, the present invention has the following beneficial effects:
[0045] 1. Spatiotemporal graph neural networks can effectively process complex patterns and nonlinear relationships in solar activity data. Solar activity is affected by many factors, including internal physical processes of the sun and external environmental factors. There are complex interactions between these factors. Traditional statistical and mathematical models may not be able to capture these complex relationships. Neural networks can make accurate predictions by learning patterns and features in the data.
[0046] 2. The space-time graph neural network can comprehensively consider the space-time information, that is, considering the changes in solar activity in time and space, it can better capture the dynamic characteristics of solar activity, including the formation, evolution and disappearance of sunspots. Traditional prediction methods may tend to consider the periodic characteristics in time and ignore the changes in space, while the space-time graph neural network can consider these two aspects of information more comprehensively.
[0047] 3. Neural networks have the ability to automatically extract features and can learn the most predictive features from raw data. There is no need for manual selection and extraction of features, which reduces the impact of human factors on prediction results and can better adapt to changes and complexity of data.
[0048] 4. The spatiotemporal graph neural network has certain adaptability and flexibility. It can be adjusted and optimized according to different data and tasks. It can also design customized models according to the characteristics of solar activity data and prediction requirements, thereby improving the accuracy and reliability of predictions. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 It is a schematic diagram of the steps of the method of the present invention. DETAILED DESCRIPTION
[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.
[0051] See also Figure 1 The present invention provides a technical solution: a method for predicting sunspot area based on spatiotemporal graph neural network, comprising the following steps:
[0052] S1. Data collection: Obtain solar observation data from astronomical observatories, including the area, location and time information of sunspots;
[0053] S2, data processing: cleaning and preprocessing the sample data obtained in step S1, and then integrating the processed data to mine the associated solar state information to form multiple related data packets;
[0054] It should be noted that the processing of data includes:
[0055] Data cleaning: the process of checking and processing errors, missing values, outliers and duplicate values in the acquired solar observation data. Data cleaning techniques include removing duplicate values, filling missing values, processing outliers and screening useful data.
[0056] Data conversion: convert solar observation data from raw formats to formats more suitable for analysis. Data conversion techniques include data encoding, format conversion, data normalization, and aggregation.
[0057] Data integration: Combining solar observation data from different sources and formats into a unified data set. Data integration techniques include data merging, joining and correlation.
[0058] Feature engineering: Extract, construct and select appropriate features from raw data based on the domain knowledge of solar observation data and problem requirements to improve the performance and effect of the model. Feature engineering techniques include feature extraction, feature conversion and feature selection;
[0059] Data reduction: When processing large-scale solar observation data, it is necessary to reduce the size and complexity of the data and improve computational efficiency. Data reduction techniques include data sampling, downsampling, and dimensionality reduction.
[0060] Data standardization: Convert solar observation data to the same scale and range to eliminate differences between data. Data standardization techniques include normalization and standardization.
[0061] Data Visualization: Use charts and graphic visualization tools to display and present solar observation data to help understand the distribution, relationships and trends of the data. Data visualization techniques include bar charts, line charts, scatter plots and heat maps.
[0062] S3, constructing a space-time map: constructing a space-time map of sunspots based on observation data;
[0063] It should be noted that constructing a space-time graph includes:
[0064] S31. Define nodes and edges: each sunspot is regarded as a node in the graph. The characteristics of the node are the area and position of the sunspot. According to the time relationship between sunspots, define edges to connect the nodes. The connection of the edges is determined by calculating the time interval between sunspots or other correlation indicators.
[0065] S32. Construction of space-time graph: Based on the defined nodes and edges, a space-time graph structure of sunspots is constructed. A graph data structure or a network data structure is used to represent the space-time graph. Each node represents the characteristics of a sunspot, and the edges represent the temporal relationship between sunspots.
[0066] S33, feature extraction and encoding: for each node and edge, feature extraction and encoding are performed;
[0067] S34. Representation and storage of space-time graph: The constructed space-time graph is represented as an adjacency matrix and stored in an appropriate data structure.
[0068] The adjacency matrix is a way to represent the connection relationship between nodes in the graph data structure. The methods for obtaining the adjacency matrix include the adjacency list representation (Adjacency List) and the adjacency matrix representation (Adjacency Matrix)
[0069] Adjacency List: In adjacency list representation, each node has a list of nodes adjacent to it. The steps to obtain the adjacency matrix include:
[0070] ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0;
[0071] ②Traverse each node in the graph, and for each node, find a list of nodes adjacent to it;
[0072] ③ Set the elements of the corresponding rows and columns in the matrix to 1, indicating that there is a connection relationship between the two nodes.
[0073] Adjacency Matrix: In the adjacency matrix representation, a two-dimensional matrix is used to represent the connection relationship between nodes in the graph. The steps to obtain the adjacency matrix include:
[0074] ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0;
[0075] ② Traverse each edge in the graph, and for each edge, set the corresponding element in the matrix to 1, indicating that there is a connection relationship between the two nodes;
[0076] ③If the graph is a weighted graph, the elements in the matrix can be set to the weight values of the edges, indicating the strength of the connection between the nodes
[0077] S4. Space-time graph neural network model training: Use the space-time graph neural network model to train the constructed space-time graph. The model will learn the space-time relationship of sunspots and model it with the sunspot area.
[0078] It should be noted that by using a graph convolutional neural network to train the constructed space-time graph, the model will learn the space-time relationship of sunspots and model it with the sunspot area;
[0079] Graph Convolutional Neural Network (GCN) is a deep learning model for processing graph structured data. In GCN, the relationship between nodes is represented by the adjacency matrix of the graph, and the graph convolution layer updates and propagates node features by performing convolution operations on the adjacency matrix.
[0080] The formula for GCN is:
[0081] [H^{(l+1)}=f(\tilde{D}^{-\frac{1}{2}}\tilde{A}\tilde{D}^{-\fra c{1}{2}}H^{(l)}W^{(l)})]
[0082] in:
[0083] (H^{(l)}) is the node feature matrix of the (l)th layer, and (H^{(0)}) is the input node feature matrix.
[0084] (\tilde{A}) is the modified adjacency matrix of the graph (A) with self-connection added (usually with the identity matrix).
[0085] (\tilde{D}) is the degree matrix of (\tilde{A}).
[0086] (W^{(l)}) is the weight matrix of the (l)th layer.
[0087] (f) is the activation function, usually a nonlinear function such as ReLU;
[0088] This formula describes the calculation process of one layer in GCN. First, the node feature matrix (H^{(l)}) is multiplied by the weight matrix (W^{(l)}) to obtain the linear transformation result of the node. Then, the information of neighbor nodes is propagated through the adjacency matrix (\tilde{A}), and the activation function (f) is used to introduce nonlinearity, and finally the node feature matrix (H^{(l+1)}) of the (l+1)th layer is obtained.
[0089] The training process of the spatiotemporal graph neural network model includes:
[0090] S41, data preparation: extract and prepare training data, including the spatiotemporal map data of sunspots and the corresponding sunspot areas;
[0091] S42. Model design: Design the structure of the spatiotemporal graph neural network model, including the model's hierarchy and connection methods;
[0092] S43, model initialization: initializing the weight parameters of the spatiotemporal graph neural network model;
[0093] S44, forward propagation: forward propagation of training data through the spatiotemporal graph neural network model to generate prediction results;
[0094] S45. Loss function definition: Define a loss function to measure the difference between the predicted result and the true label. The loss function includes mean square error (MSE) and cross entropy loss function.
[0095] S46, Back Propagation: Use the back propagation algorithm to calculate the gradient of the model parameters, and update the model parameters according to the gradient to minimize the loss function. The optimization algorithms include Stochastic Gradient Descent (SGD) and Adam.
[0096] S47, iterative training: through multiple iterations, repeat the forward propagation and back propagation steps, and use the training data to continuously optimize the model parameters so that the model can better fit the data
[0097] S5. Predict sunspot area: Use the trained space-time graph neural network model to predict the future sunspot space-time graph.
[0098] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A sunspot area prediction method based on spatiotemporal graph neural network, characterized by: The following steps are involved: S1. Data collection: Obtain solar observation data from astronomical observatories, including the area, location and time information of sunspots; S2, data processing: cleaning and preprocessing the sample data obtained in step S1, and then integrating the processed data to mine the associated solar state information to form multiple related data packets; S3, constructing a space-time map: constructing a space-time map of sunspots based on observation data; S4. Space-time graph neural network model training: Use the space-time graph neural network model to train the constructed space-time graph. The model will learn the space-time relationship of sunspots and model it with the sunspot area. S5. Predict sunspot area: Use the trained space-time graph neural network model to predict the future sunspot space-time graph.
2. The method for predicting sunspot area based on spatiotemporal graph neural network according to claim 1, characterized in that: The data processing in step S2 includes: Data cleaning: the process of checking and processing errors, missing values, outliers and duplicate values in the acquired solar observation data. Data cleaning techniques include removing duplicate values, filling missing values, processing outliers and screening useful data. Data conversion: convert solar observation data from raw formats to formats more suitable for analysis. Data conversion techniques include data encoding, format conversion, data normalization, and aggregation. Data integration: Combining solar observation data from different sources and formats into a unified data set. Data integration techniques include data merging, joining and correlation. Feature engineering: Extract, construct and select appropriate features from raw data based on the domain knowledge of solar observation data and problem requirements to improve the performance and effect of the model. Feature engineering techniques include feature extraction, feature conversion and feature selection; Data reduction: When processing large-scale solar observation data, it is necessary to reduce the size and complexity of the data and improve computational efficiency. Data reduction techniques include data sampling, downsampling, and dimensionality reduction. Data standardization: Convert solar observation data to the same scale and range to eliminate differences between data. Data standardization techniques include normalization and standardization. Data Visualization: Use charts and graphic visualization tools to display and present solar observation data to help understand the distribution, relationships and trends of the data. Data visualization techniques include bar charts, line charts, scatter plots and heat maps.
3. The sunspot area prediction method based on spatiotemporal graph neural network according to claim 1 is characterized in that: The step S3 of constructing the space-time graph includes: S31. Define nodes and edges: each sunspot is regarded as a node in the graph. The characteristics of the node are the area and position of the sunspot. According to the time relationship between sunspots, define edges to connect the nodes. The connection of the edges is determined by calculating the time interval between sunspots or other correlation indicators. S32. Construction of space-time graph: Based on the defined nodes and edges, a space-time graph structure of sunspots is constructed. A graph data structure or a network data structure is used to represent the space-time graph. Each node represents the characteristics of a sunspot, and the edges represent the temporal relationship between sunspots. S33, feature extraction and encoding: for each node and edge, feature extraction and encoding are performed; S34. Representation and storage of space-time graph: The constructed space-time graph is represented as an adjacency matrix and stored in an appropriate data structure.
4. The method for predicting sunspot area based on spatiotemporal graph neural network according to claim 3, characterized in that: The adjacency matrix is a way to represent the connection relationship between nodes in a graph data structure. The methods for obtaining the adjacency matrix include the adjacency list representation method (Adjacency List) and the adjacency matrix representation method (Adjacency Matrix) Adjacency List: In adjacency list representation, each node has a list of nodes adjacent to it. The steps to obtain the adjacency matrix include: ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0; ②Traverse each node in the graph, and for each node, find a list of nodes adjacent to it; ③ Set the elements of the corresponding rows and columns in the matrix to 1, indicating that there is a connection relationship between the two nodes. Adjacency Matrix: In the adjacency matrix representation, a two-dimensional matrix is used to represent the connection relationship between nodes in the graph. The steps to obtain the adjacency matrix include: ① Create a two-dimensional matrix with a size of the number of nodes * the number of nodes, and initialize all elements to 0; ② Traverse each edge in the graph, and for each edge, set the corresponding element in the matrix to 1, indicating that there is a connection relationship between the two nodes; ③If the graph is a weighted graph, the elements in the matrix can be set to the weight values of the edges, indicating the strength of the connection between the nodes.
5. The method for predicting sunspot area based on spatiotemporal graph neural network according to claim 1, characterized in that: In step S4, a graph convolutional neural network is used to train the constructed space-time graph, and the model learns the space-time relationship of sunspots and models it with the sunspot area.
6. The method for predicting sunspot area based on spatiotemporal graph neural network according to claim 1, characterized in that: The spatiotemporal graph neural network model training process in step S4 includes: S41, data preparation: extract and prepare training data, including the spatiotemporal map data of sunspots and the corresponding sunspot areas; S42. Model design: Design the structure of the spatiotemporal graph neural network model, including the model's hierarchy and connection methods; S43, model initialization: initializing the weight parameters of the spatiotemporal graph neural network model; S44, forward propagation: forward propagation of training data through the spatiotemporal graph neural network model to generate prediction results; S45. Loss function definition: Define a loss function to measure the difference between the predicted result and the true label. The loss function includes mean square error (MSE) and cross entropy loss function. S46, Back Propagation: Use the back propagation algorithm to calculate the gradient of the model parameters, and update the model parameters according to the gradient to minimize the loss function. The optimization algorithms include Stochastic Gradient Descent (SGD) and Adam. S47, Iterative training: Through multiple iterations, repeat the forward propagation and back propagation steps, and use the training data to continuously optimize the model parameters so that the model can better fit the data.