A Method for Constructing a Digital Twin Model of Satellite Temperature Field Based on Graph Neural Networks

Abstract

This invention discloses a method for constructing a digital twin model of a satellite temperature field based on a graph neural network, comprising: determining a first type of location points with temperature sensors and a second type of location points without sensors; constructing a graph model based on the distance relationships between the location points; acquiring multiple sets of historical data; generating a training dataset based on the historical data and the graph model; constructing a graph convolutional neural network and training the graph convolutional neural network using the training dataset; obtaining the temperature of the second type of location points using the graph convolutional neural network based on the temperature data of the first type of location points collected by the sensors; visualizing the temperatures of the first type of location points and the second type of location points to obtain a digital twin model of the satellite temperature field. This invention can obtain the temperatures of all temperature-sensitive points using the temperatures measured at a finite number of temperature-sensitive points, thereby obtaining the corresponding digital twin model of the satellite temperature field, and is applicable to unstructured two-dimensional or three-dimensional scenes.

Description

Technical Field

[0001] This invention relates to the field of satellite engineering technology, specifically to a method for constructing a digital twin model of a satellite temperature field based on graph neural networks. Background Technology

[0002] Satellites in orbit need to undertake various experimental and engineering tasks, and also need to control their attitude and velocity. Therefore, they carry a large number of electronic devices. The heat-generating components of these electronic devices are densely distributed, resulting in a high heat source density. Furthermore, the temperature in space varies greatly, and electronic devices are extremely sensitive to temperature changes. If the operating temperature exceeds the range that the device can withstand, it can cause the device to shut down or even be damaged, potentially jeopardizing the normal operation of the satellite or even causing it to shut down or be damaged. This would be a catastrophic blow to a satellite in orbit.

[0003] Due to limitations in space and payload costs, satellites cannot deploy sensors at all temperature-sensitive points to monitor temperature in real time. They can only deploy sensors at some temperature-sensitive points for temperature measurement. Therefore, they can only conduct real-time prediction and monitoring of the internal temperature field of the satellite by establishing a digital twin based on limited sensor information.

[0004] There are currently two main approaches to real-time prediction and monitoring of the internal temperature field of a satellite based on measurements from multiple sensors. The first approach involves calculating the satellite's temperature field using traditional numerical methods based on the sensor data, and then determining the temperature of other temperature-sensitive points based on this predicted field. The second approach utilizes convolutional neural networks to construct a digital twin surrogate model. The sensor data is used as input to this surrogate model, which then predicts and outputs the corresponding satellite temperature field. Finally, the predicted satellite temperature field is used to determine the temperature of other temperature-sensitive points.

[0005] However, traditional numerical computation methods require multiple iterations, resulting in low computational efficiency and difficulty in achieving real-time performance. This fails to fully realize the application value of digital twins and also encounters computational bottlenecks when dealing with high-dimensional problems, easily leading to the "curse of dimensionality." Existing methods using convolutional neural networks to construct surrogate models for digital twins treat temperature fields as images, limiting their application to regular two-dimensional scenarios and making it difficult to solve unstructured two-dimensional or three-dimensional problems, thus exhibiting certain limitations in practical applications. Summary of the Invention

[0006] In order to solve some or all of the technical problems existing in the prior art, the present invention provides a method for constructing a digital twin model of satellite temperature field based on graph neural network.

[0007] The technical solution of the present invention is as follows:

[0008] A method for constructing a digital twin model of a satellite temperature field based on a graph neural network is provided, the method comprising:

[0009] Identify the first type of location points on the satellite where temperature sensors are deployed, and the second type of location points where temperature monitoring is required but no temperature sensors are deployed.

[0010] Treat each location point as a node, and construct a graph model based on the distance relationships between various location points;

[0011] Multiple sets of historical data are acquired, wherein each set of historical data includes: the temperature of the first type of location point, the power information of the heat-generating components inside the satellite, the satellite cabin operating condition information, and the temperature of the second type of location point;

[0012] Based on the historical data and the graph model, a training dataset is generated that includes multiple training data, wherein the training data includes a graph model containing node input features and the temperature of the corresponding second type of location points;

[0013] Construct a graph convolutional neural network and train the graph convolutional neural network using the training dataset to fit the mapping relationship between the input data and the temperature of the second type of location points, wherein the input data is a graph model containing node input features;

[0014] Based on the temperature data of the first type of location points collected by the temperature sensor, the power information of the heating components inside the satellite, and the operating condition information of the satellite cabin, the corresponding graph model containing node input features is determined. The graph model containing node input features is then input into the trained graph convolutional neural network. The temperatures of the first type of location points collected by the temperature sensor and the temperatures of the second type of location points output by the graph convolutional neural network are visualized to obtain a digital twin model of the satellite temperature field.

[0015] In some possible implementations, a location point is treated as a node, and a graph model is constructed based on the distance relationships between the various location points, including:

[0016] Treat a location point as a node, and determine the nodes corresponding to all location points;

[0017] Determine the distance between each location point. If the distance between two location points is less than a preset distance threshold, add an undirected edge between the two nodes corresponding to the two location points.

[0018] Construct the corresponding graph model based on all the identified nodes and edges.

[0019] In some possible implementations, the satellite cabin operating condition information includes: the temperature of the sunlit side of the satellite and / or the temperature of the shaded side of the satellite.

[0020] In some possible implementations, a training dataset comprising multiple training data is generated based on the historical data and the graph model, including:

[0021] The temperature of the first type of location point, the power information of the heat-generating components inside the satellite, and the operating condition information of the satellite cabin are processed into the input features of the nodes corresponding to the first type of location point in the graph model. The graph model containing the node input features is used as the training sample, and the temperature of the corresponding second type of location point is used as the sample label. Multiple training data including the training samples and their corresponding sample labels are obtained to generate a training dataset including multiple training data.

[0022] In some possible implementations, one-dimensional convolution is used to process the temperature of the first type of location points, the power information of the heat-generating components inside the satellite, and the operating conditions of the satellite cabin. The processed data is then spliced ​​and fused, and the spliced ​​and fused data is used as the input features of the nodes.

[0023] In some possible implementations, batch normalization strategies and residual connections are added to the constructed graph convolutional neural network.

[0024] In some possible implementations, training a graph convolutional neural network using the training dataset includes:

[0025] The graph model containing node input features in the training data is used as the input to the graph convolutional neural network, and the temperature of the second type of location points in the training data is used as the output of the graph convolutional neural network to train the graph convolutional neural network.

[0026] In some possible implementations, the graph model containing node input features in the training data is used as the input to the graph convolutional neural network, and the temperature of the second type of location points in the training data is used as the output of the graph convolutional neural network. Training the graph convolutional neural network includes:

[0027] Step S51: Input the graph models containing node input features from multiple training data into the graph convolutional neural network in sequence to obtain the temperature prediction values ​​of the second type of location points output by the graph convolutional neural network.

[0028] Step S52: Compare the predicted temperature values ​​of the second type of location points with the temperatures of the second type of location points in the training data, and calculate the prediction accuracy of the graph convolutional neural network.

[0029] Step S53: Determine whether the prediction accuracy obtained in at least two consecutive steps is greater than the preset accuracy threshold. If yes, use the current graph convolutional neural network as the trained graph convolutional neural network. If no, calculate the loss function, update the parameters of the graph convolutional neural network using the loss function, and return to step S51.

[0030] In some possible implementations, the loss function is:

[0031]

[0032] Where N represents the number of training data, y i This represents the temperature of the second type of location point in the i-th training data. This represents the temperature prediction value of the second type of location point output by the graph convolutional neural network corresponding to the i-th training data.

[0033] In some possible implementations, the method further includes:

[0034] The acquired digital twin model of the satellite temperature field will be transmitted to other systems on the satellite.

[0035] The main advantages of the technical solution of this invention are as follows:

[0036] The satellite temperature field digital twin model construction method based on graph neural network of the present invention can obtain the temperature of all temperature sensitive points by using the temperature measured by a limited number of temperature sensitive points, realize real-time temperature monitoring of all temperature sensitive points inside the satellite, and obtain the corresponding satellite temperature field digital twin model. It can also fully consider the working conditions of the internal components of the satellite and is applicable to unstructured two-dimensional or three-dimensional scenarios. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a method for constructing a satellite temperature field digital twin model based on a graph neural network, according to an embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] The technical solutions provided by the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] refer to Figure 1 An embodiment of the present invention provides a method for constructing a digital twin model of a satellite temperature field based on a graph neural network, the method comprising the following steps:

[0042] Step S1: Determine the first type of location points on the satellite where temperature sensors are deployed, and the second type of location points where temperature monitoring is required but no temperature sensors are deployed.

[0043] Step S2: Treat a location point as a node and construct a graph model based on the distance relationships between the various location points;

[0044] Step S3: Obtain multiple sets of historical data, wherein each set of historical data includes: the temperature of the first type of location point, the power information of the heat-generating components inside the satellite, the satellite cabin operating condition information, and the temperature of the second type of location point;

[0045] Step S4: Based on historical data and the graph model, generate a training dataset that includes multiple training data, wherein the training data includes a graph model containing node input features and the temperature of the corresponding second type of location points;

[0046] Step S5: Construct a graph convolutional neural network. Use the training dataset to train the graph convolutional neural network to fit the mapping relationship between the graph model containing node input features and the temperature of the second type of location points.

[0047] Step S6: Based on the temperature data of the first type of location points collected by the temperature sensor, the power information of the heating components inside the satellite, and the operating condition information of the satellite cabin, determine the corresponding graph model containing node input features, input the graph model containing node input features into the trained graph convolutional neural network, and perform visualization processing on the temperature of the first type of location points collected by the temperature sensor and the temperature of the second type of location points output by the graph convolutional neural network to obtain a digital twin model of the satellite temperature field.

[0048] The following provides a detailed explanation of the steps and principles of a satellite temperature field digital twin model construction method based on graph neural networks, according to an embodiment of the present invention:

[0049] Step S1: Determine the first type of location points on the satellite where temperature sensors are deployed, and the second type of location points where temperature monitoring is required but no temperature sensors are deployed.

[0050] Due to space and payload cost constraints, satellites cannot deploy sensors at all temperature-sensitive points for real-time temperature monitoring; they can only deploy sensors at select points. Therefore, in practical applications, a certain number of temperature-sensitive points can be selected for temperature sensor deployment. These points with deployed sensors are designated as Category I location points, while those without are designated as Category II location points.

[0051] Step S2: Treat a location point as a node and construct a graph model based on the distance relationships between the various location points.

[0052] Graph models are unstructured data structures capable of storing and processing irregular data. The key to constructing a graph model is determining its nodes and edges. A graph G = (V, E) is defined, where V represents the set of nodes, a collection of entity nodes that stores node characteristics; E represents the set of edges, describing the connections between nodes, with edge weights representing different connections. To better describe the connections between nodes and edges, an adjacency matrix A and a degree matrix D are typically used. The adjacency matrix represents the adjacency relationships between nodes, and the degree matrix is ​​a matrix composed of the degrees of each node, where the degree represents the number of edges associated with that node.

[0053] In one embodiment of the present invention, a location point is taken as a node, and a graph model is constructed based on the distance relationship between the various location points.

[0054] Specifically, taking a location point as a node, a graph model is constructed based on the distance relationships between various location points, including the following steps:

[0055] Treat a location point as a node, and determine the nodes corresponding to all location points;

[0056] Determine the distance between each location point. If the distance between two location points is less than a preset distance threshold, add an undirected edge between the two nodes corresponding to the two location points.

[0057] Construct the corresponding graph model based on all the identified nodes and edges.

[0058] The preset distance threshold can be specifically set according to the internal component structure, component layout, heat dissipation method, and the influence range of component temperature.

[0059] Step S3: Obtain multiple sets of historical data.

[0060] Specifically, multiple sets of historical data can be collected through various means, including the temperature of the first type of location point, the power information of the heat-generating components inside the satellite, the satellite cabin operating condition information, and the temperature of the second type of location point. Among them, the temperature of the first type of location point, the power information of the heat-generating components inside the satellite, the satellite cabin operating condition information, and the temperature of the second type of location point in the same set of historical data represent data of the satellite in the same state; different sets of historical data correspond to different satellite states.

[0061] In one embodiment of the present invention, the historical data can be real data or historical simulation data.

[0062] In one embodiment of the present invention, the satellite cabin operating condition information may include: the temperature of the sun-facing side of the satellite and / or the temperature of the shaded side of the satellite.

[0063] Step S4: Generate a training dataset that includes multiple training data based on historical data and the graph model.

[0064] Specifically, in one embodiment of the present invention, a training dataset including multiple training data is generated based on historical data and a graph model, comprising the following steps:

[0065] The temperature of the first type of location point, the power information of the heat-generating components inside the satellite, and the operating condition information of the satellite cabin are processed into the input features of the nodes corresponding to the first type of location point in the graph model. The graph model containing the node input features is used as the training sample, and the temperature of the corresponding second type of location point is used as the sample label. Multiple training data including the training samples and their corresponding sample labels are obtained to generate a training dataset including multiple training data.

[0066] Furthermore, considering that the dimensions of the temperature at the first type of location point, the power information of the heating components inside the satellite, and the operational status information of the satellite cabin may differ, in one embodiment of the present invention, one-dimensional convolution is used to process the temperature at the first type of location point, the power information of the heating components inside the satellite, and the operational status information of the satellite cabin. The processed data is then spliced ​​and fused, and the spliced ​​and fused data is used as the input features of the nodes.

[0067] By using one-dimensional convolution to process information data, information of different dimensions can be expanded to features with the same number of dimensions and nodes, thus achieving standardization of information data. Then, by splicing and fusing the processed information data, the temperature of the first type of location point, the power information of the heating components inside the satellite, and the operating conditions of the satellite cabin can be processed into node features that can be used by graph convolutional neural networks.

[0068] Step S5: Construct a graph convolutional neural network. Use the training dataset to train the graph convolutional neural network to fit the mapping relationship between the graph model containing node input features and the temperature of the second type of location points.

[0069] Graph neural networks (GNNs) introduce deep learning into graph theory to handle problems involving irregular regions. GNNs are a type of neural network that applies neural networks to graph processing, enabling them to extract neighborhood information from nodes and pass it to the next layer via learnable parameters.

[0070] In one embodiment of the present invention, a graph convolutional network (GCN) is used as the neural network for feature extraction.

[0071] Graph convolutional neural networks (GCNNs) consist of multiple graph convolutional layers. Through graph convolution operations, GCNNs can extract and aggregate neighborhood information to obtain hidden layer representations of nodes. Graph convolution operations mainly consist of two steps: aggregation, which involves aggregating information from neighboring nodes, and propagation, which involves passing the node information to subsequent layers. Specifically, this can be described as follows:

[0072]

[0073] Among them, H l+1 Let σ(·) represent the features of the (l+1)th layer of the graph convolutional neural network. Let A represent the adjacency matrix with a self-loop. Let D represent the degree matrix with self-loops, and I represent the degree matrix. N H represents the identity matrix. l θ represents the features of the l-th layer of the graph convolutional neural network. l H0 represents the parameters of the l-th layer of the graph convolutional neural network; when l = 0, H0 represents the input of the graph convolutional neural network; when l = K, K is the number of layers in the graph convolutional neural network, H0... K This represents the output of the graph convolutional neural network.

[0074] The adjacency matrix is ​​used to represent the interconnections between nodes. If there is an undirected edge between node i and node j, then the corresponding adjacency matrix element A... ij =1, otherwise A ij =0; The degree matrix is ​​used to represent the number of edges for each node, and is a diagonal matrix. M represents the number of columns in the adjacency matrix.

[0075] Furthermore, in one embodiment of the present invention, in order to improve prediction accuracy and avoid the graph convolutional neural network from falling into the problem of oversmoothing, a batch normalization strategy and a residual connection are added to the graph convolutional neural network.

[0076] Specifically, after incorporating batch normalization and the form of residual connections, the graph convolution operation can be represented as:

[0077]

[0078] Where BN(·) represents batch normalization.

[0079] Furthermore, after constructing the graph convolutional neural network, in order to improve the prediction accuracy of the graph convolutional neural network and reduce the prediction error, it is necessary to train and update the graph convolutional neural network using training data.

[0080] In one embodiment of the present invention, training a graph convolutional neural network using a training dataset includes:

[0081] The graph model containing node input features in the training data is used as the input to the graph convolutional neural network, and the temperature of the second type of location points in the training data is used as the output of the graph convolutional neural network to train the graph convolutional neural network.

[0082] Specifically, in one embodiment of the present invention, the graph model containing node input features in the training data is used as the input of the graph convolutional neural network, and the temperature of the second type of location points in the training data is used as the output of the graph convolutional neural network. Training the graph convolutional neural network further includes the following steps:

[0083] Step S51: Input the graph models containing node input features from multiple training data into the graph convolutional neural network in sequence to obtain the temperature prediction values ​​of the second type of location points output by the graph convolutional neural network.

[0084] In one embodiment of the present invention, a graph model containing node input features from the training data is input into the input end of a graph convolutional neural network (Graph Convolutional Neural Network). The model is processed sequentially through the parameters of each layer in the Graph Convolutional Neural Network and output from the output end. The information output is the predicted temperature value of the second type of location point corresponding to the graph model containing node input features. The initial Graph Convolutional Neural Network can be an untrained neural network or a neural network that has not been fully trained. Each layer has initialized parameters, which can be continuously updated and adjusted during the training process of the neural network.

[0085] Step S52: Compare the predicted temperature values ​​of the second type of location points with the temperatures of the second type of location points in the training data, and calculate the prediction accuracy of the graph convolutional neural network.

[0086] In one embodiment of the present invention, the average of all ratios can be used as the prediction accuracy by calculating the difference between the predicted temperature value of the second type of location point corresponding to each training data point and the temperature of the second type of location point and the temperature of the second type of location point.

[0087] Step S53: Determine whether the prediction accuracy obtained in at least two consecutive steps is greater than the preset accuracy threshold. If yes, use the current graph convolutional neural network as the trained graph convolutional neural network. If no, calculate the loss function, update the parameters of the graph convolutional neural network using the loss function, and return to step S51.

[0088] In one embodiment of the present invention, the mean absolute error can be used as the loss function of the graph convolutional neural network. In this case, the loss function can be specifically expressed as:

[0089]

[0090] Where N represents the number of training data, y i This represents the temperature of the second type of location point in the i-th training data. This represents the temperature prediction value of the second type of location point output by the graph convolutional neural network corresponding to the i-th training data. The number of training data N can be determined according to the actual training requirements.

[0091] In one embodiment of the present invention, gradient descent is used to optimize and update the parameters of a graph convolutional neural network. Specifically, the chain rule is first used to differentiate the parameters with respect to the loss function, and then the parameters are updated using the differentiation result and a preset learning rate.

[0092] Specifically, the parameters of a graph convolutional neural network can be updated using the following formula:

[0093]

[0094] Where θ represents the set of parameters of the graph convolutional neural network, and η represents the learning rate, which needs to be set in advance to control how fast the parameters are updated.

[0095] Step S6: Based on the temperature data of the first type of location points collected by the temperature sensor, the power information of the heating components inside the satellite, and the operating condition information of the satellite cabin, determine the corresponding graph model containing node input features, input the graph model containing node input features into the trained graph convolutional neural network, and perform visualization processing on the temperature of the first type of location points collected by the temperature sensor and the temperature of the second type of location points output by the graph convolutional neural network to obtain a digital twin model of the satellite temperature field.

[0096] In one embodiment of the present invention, after training the graph convolutional neural network, the temperature data of the first type of location points collected in real time by multiple temperature sensors, as well as the power information of the heat-generating components inside the satellite and the operating condition information of the satellite cabin at the same time, are processed and fused to determine the input features of the nodes. The graph model containing the node input features is input into the graph convolutional neural network, and the graph convolutional neural network outputs the temperature of the corresponding second type of location point. By visualizing the temperature of the first type of location points collected in real time by the temperature sensors and the temperature of the second type of location points output in real time by the graph convolutional neural network, a digital twin model of the satellite temperature field is obtained.

[0097] Furthermore, in one embodiment of the present invention, the method further includes:

[0098] The acquired digital twin model of the satellite temperature field will be transmitted to other systems on the satellite.

[0099] By transmitting the satellite temperature field digital twin model to other systems within the satellite, information exchange can be achieved between multiple systems and multiple digital twins within the satellite, facilitating satellite control and management.

[0100] The satellite temperature field digital twin model construction method based on graph neural network provided in one embodiment of the present invention can obtain the temperature of all temperature sensitive points by using the temperature measured by a limited number of temperature sensitive points, realize real-time temperature monitoring of all temperature sensitive points inside the satellite, and obtain the corresponding satellite temperature field digital twin model. It can also fully consider the working conditions of the internal components of the satellite and is applicable to unstructured two-dimensional or three-dimensional scenarios.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Additionally, the terms "front," "back," "left," "right," "upper," and "lower" in this document refer to the placement shown in the accompanying drawings.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A satellite temperature field digital twin model construction method based on a graph neural network, characterized in that, The method comprises the following steps: determining a first type of position points on the satellite where temperature sensors are arranged, and a second type of position points where temperature monitoring is required but no temperature sensor is arranged; taking a position point as a node, and constructing a graph model according to the distance relationship between the position points; obtaining a plurality of sets of historical data, wherein each set of historical data comprises the temperature of the first type of position points, the power information of the heat generating components in the satellite, the satellite cabin working condition information, and the temperature of the second type of position points; generating a training data set comprising a plurality of training data according to the historical data and the graph model, wherein the training data comprises a graph model comprising node input features and the temperature of the corresponding second type of position points; constructing a graph convolutional neural network, and training the graph convolutional neural network by using the training data set to fit the mapping relationship between the input data and the temperature of the second type of position points, wherein the input data is the graph model comprising node input features; determining the corresponding graph model comprising node input features according to the temperature data of the first type of position points collected by the temperature sensor, the power information of the heat generating components in the satellite, and the satellite cabin working condition information, inputting the graph model comprising node input features into the trained graph convolutional neural network, and performing visual processing on the temperature of the first type of position points collected by the temperature sensor and the temperature of the second type of position points output by the graph convolutional neural network to obtain a satellite temperature field digital twin model.

2. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 1, characterized in that, The method of taking a position point as a node and constructing a graph model according to the distance relationship between the position points comprises the following steps: taking a position point as a node, and determining the nodes corresponding to all the position points; determining the distance between each position point, and adding an undirected edge between the nodes corresponding to two position points if the distance between the two position points is less than a preset distance threshold; constructing a corresponding graph model according to all the determined nodes and edges.

3. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 1, characterized in that, The satellite cabin working condition information comprises the temperature of the sun-facing surface of the satellite and / or the temperature of the shadow surface of the satellite.

4. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 1, characterized in that, The method of generating a training data set comprising a plurality of training data according to the historical data and the graph model comprises the following steps: processing the temperature of the first type of position points, the power information of the heat generating components in the satellite, and the satellite cabin working condition information into the input features of the nodes corresponding to the first type of position points in the graph model, taking the graph model comprising node input features as a training sample, taking the temperature of the corresponding second type of position points as a sample label, obtaining a plurality of training data comprising the training sample and the corresponding sample label, and generating a training data set comprising a plurality of training data.

5. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 4, characterized in that, processing the temperature of the first type of position points, the power information of the heat generating components in the satellite, and the satellite cabin working condition information by using one-dimensional convolution, splicing and fusing the data obtained after processing, and taking the spliced and fused data as the input features of the nodes.

6. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 1, characterized in that, adding a batch normalization strategy and a residual connection in the constructed graph convolutional neural network.

7. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 1, characterized in that, The method of training the graph convolutional neural network by using the training data set comprises the following steps: taking the graph model comprising node input features in the training data as the input of the graph convolutional neural network, taking the temperature of the second type of position points in the training data as the output of the graph convolutional neural network, and training the graph convolutional neural network.

8. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 7, characterized in that, The graph model containing node input features in the training data is taken as the input of the graph convolutional neural network, and the temperature of the second type of position point in the training data is taken as the output of the graph convolutional neural network, and the graph convolutional neural network is trained, comprising: In step S51, the graph model containing node input features in the plurality of training data is sequentially input into the graph convolutional neural network to obtain the temperature prediction value of the second type of position point output by the graph convolutional neural network; In step S52, the temperature prediction value of the second type of position point is compared with the temperature of the second type of position point in the training data, and the prediction accuracy of the graph convolutional neural network is calculated; In step S53, it is judged whether the prediction accuracy obtained at least twice continuously is greater than a preset accuracy threshold value, if yes, the current graph convolutional neural network is taken as the trained graph convolutional neural network, and if no, a loss function is calculated, the parameters of the graph convolutional neural network are updated by using the loss function, and the step S51 is returned.

9. The satellite temperature field digital twin model construction method based on a graph neural network according to claim 8, characterized in that, The loss function is: ； wherein N represents the number of training data, represents the temperature of the second type position point in the i-th training data, represents the temperature prediction value of the second type position point output by the graph convolutional neural network corresponding to the i-th training data.

10. The graph neural network-based satellite temperature field digital twin model construction method according to any one of claims 1-9, characterized in that, The method further comprises: The obtained satellite temperature field digital twin model is transmitted to other systems of the satellite.

Images