Proximity event prediction method and device, electronic equipment and medium

Abstract

The invention discloses a proximity event prediction method and device, electronic equipment and a medium, and the method comprises the steps: constructing a training data set related to satellite proximity events, and constructing a satellite knowledge graph; based on the satellite knowledge graph, predicting whether a proximity event exists between any two satellite nodes, and obtaining a proximity prediction result; inputting the training data set into a pre-constructed graph neural network for training to obtain a proximity training result; constructing a loss function through a proximity prediction result and a proximity training result; performing iterative training by taking loss function minimization as a target to obtain a target map neural network; and inputting the feature data of the to-be-predicted satellite into the target map neural network to obtain a target proximity prediction result. Therefore, the loss function constructed by the proximity prediction result and the proximity training result is minimized, training of the graph neural network is completed, and the efficient and accurate target graph neural network which can be directly applied to proximity event prediction is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aerospace engineering, and in particular to a near approach event prediction method and device, electronic equipment and a medium. BACKGROUND

[0002] With the rapid development of aerospace technology, the number of satellites has shown explosive growth, making the space environment of low earth orbit increasingly complex and crowded. At present, a large number of satellites and space debris are operating in a limited orbital space, resulting in a significant increase in satellite orbital density and increasingly complex relative motion relationships between satellites and between satellites and space debris.

[0003] As a result, the frequency of satellite near approach events has increased significantly, where a near approach event refers to a phenomenon in which two or more space objects reduce their relative distance to within a potential collision risk threshold due to orbital overlap or proximity during orbital operation. If timely warning and avoidance measures are not taken, a near approach event can cause a catastrophic on-orbit collision, not only causing damage to related satellites, but also generating a large number of new space debris, exacerbating the deterioration of the orbital environment and threatening the sustainable development of the orbital ecosystem.

[0004] Therefore, how to efficiently and accurately predict satellite near approach events is a problem that needs to be solved by those skilled in the art. SUMMARY

[0005] Therefore, one aspect of the present application provides a near approach event prediction method, which comprises: constructing a training data set for satellite near approach events and constructing a satellite knowledge graph; based on the satellite knowledge graph, predicting whether a near approach event exists between any two satellite nodes to obtain a near approach prediction result; inputting the training data set into a pre-constructed graph neural network for training to obtain a near approach training result; constructing a loss function through the near approach prediction result and the near approach training result; iteratively training the step of inputting the training data set into the pre-constructed graph neural network to obtain the near approach training result, with the goal of minimizing the loss function, to obtain a target graph neural network; inputting feature data of a satellite to be predicted into the target graph neural network to obtain a target near approach prediction result for the satellite to be predicted.

[0006] Optionally, the step of predicting whether a near approach event exists between any two satellite nodes based on the satellite knowledge graph to obtain a near approach prediction result comprises: extracting node features of each entity node from the satellite knowledge graph; the entity nodes include the satellite nodes and orbit nodes; mapping the node features to a unified vector space to obtain feature vectors; normalizing the feature vectors to obtain target feature vectors; inputting an element-by-element multiplication result of any two target feature vectors into a multi-layer perception mechanism to obtain a target output result; determining, according to the target output result, whether a close approach event exists between two satellite nodes to obtain the close approach prediction result.

[0007] Optionally, the extracting of the node features of each entity node from the satellite knowledge graph comprises: constructing a vertex set in a hypergraph based on the entity nodes, and constructing a hyperedge set in the hypergraph based on whether an edge exists between any two entity nodes; constructing a first adjacency matrix according to the corresponding hyperedge between any two satellite nodes, and constructing a second adjacency matrix according to the corresponding hyperedge between the satellite nodes and the orbit nodes; splicing the first adjacency matrix and the second adjacency matrix to obtain a target adjacency matrix; aggregating the features of each entity node based on the target adjacency matrix through a specified graph neural network to obtain the node features.

[0008] Optionally, the loss function is a binary cross-entropy loss function.

[0009] Optionally, the training data set further comprises a test data set, and the method comprises: inputting the test data set into the target graph neural network to obtain a target close approach prediction result; determining, according to the target close approach prediction result, a quality score for evaluating the close approach event prediction confidence of the target graph neural network; when the quality score is not greater than a threshold value, returning to the step of constructing the training data set about satellite close approach events and performing subsequent steps.

[0010] Optionally, the constructing of the satellite knowledge graph comprises: obtaining a target data set about satellite close approach events; taking satellites and orbits in the target data set as entity nodes in the satellite knowledge graph; generating edges between the entity nodes according to close approach events between satellite nodes and operating relationships between the satellite nodes and orbit nodes; injecting satellite parameters into corresponding satellite nodes and injecting orbit parameters into corresponding orbit nodes to construct the satellite knowledge graph.

[0011] Optionally, the satellite parameters at least include satellite number, payload and physical parameters, and the orbit parameters at least include orbit semi-major axis, orbit inclination and eccentricity.

[0012] Another aspect of the present application provides a prediction device for close approach events, the device comprising: a target construction module for constructing a training data set for satellite close approach events and constructing a satellite knowledge graph; a prediction module for predicting whether there is a close approach event between any two satellite nodes based on the satellite knowledge graph to obtain a close approach prediction result; a network training module for inputting the training data set into a pre-constructed graph neural network for training to obtain a close approach training result; constructing a loss function through the close approach prediction result and the close approach training result; iteratively training the step of inputting the training data set into the pre-constructed graph neural network to obtain the close approach training result with the goal of minimizing the loss function to obtain a target graph neural network; an input module for inputting feature data of a satellite to be predicted into the target graph neural network to obtain a target close approach prediction result of the satellite to be predicted.

[0013] Another aspect of the present application provides an electronic device comprising a memory and a processor, the memory having stored thereon a computer program capable of running on the processor, and the processor implements the steps of the prediction method for close approach events when executing the program.

[0014] Another aspect of the present application provides a computer-readable storage medium having stored thereon a computer program, the program being executed by a processor to implement the steps of the prediction method for close approach events.

[0015] The prediction method, device, electronic device and medium for close approach events provided by the present application have the beneficial effects that the satellite and the orbit are converted into a graph structure form of a knowledge graph, and whether there is a close approach event between satellite nodes can be quickly predicted according to the features of each node in the knowledge graph. At the same time, the graph neural network is trained through a training data set, so that the graph neural network processes complex nonlinear relationships to learn the close approach events between satellites. Further, the training of the graph neural network is completed by minimizing the loss function constructed by the close approach prediction result and the close approach training result, so that a target graph neural network which is efficient and accurate and can be directly used for close approach event prediction is obtained, thereby avoiding disastrous on-orbit collisions caused by close approach events and realizing sustainable development of the orbital environment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of a prediction method for close approach events provided by an embodiment of the present application; Figure 2 A schematic diagram of the principle of a near approach event prediction method provided by an embodiment of the present application; Figure 3 A schematic diagram of the principle of node feature extraction provided by an embodiment of the present application; Figure 4 A structural schematic diagram of a near approach event prediction device provided by an embodiment of the present application; Figure 5 A structural schematic diagram of an electronic device provided by an embodiment of the present application.

[0017] The reference signs are as follows: 40 is a target construction module, 41 is a prediction module, 42 is a network training module, 43 is an input module, 50 is a storage, 51 is a processor, 52 is a display screen, 53 is an input / output interface, 54 is a communication interface, 55 is a power supply, 56 is a communication bus, 501 is a computer program, 502 is an operating system, 503 is data. DETAILED DESCRIPTION

[0018] The terms used in the present application are merely for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a," "said," and "the" used in the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0019] It should be understood that although the terms first, second, third, etc. can be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, without departing from the scope of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "upon determining" or "in response to determining".

[0020] Figure 1 A flowchart of a near approach event prediction method provided by an embodiment of the present application is shown in Figure 1 The method comprises: S10: constructing a training data set about satellite near approach events, and constructing a satellite knowledge graph; S11: based on the satellite knowledge graph, predicting whether there is a near approach event between any two satellite nodes, to obtain a near approach prediction result; S12: inputting the training data set into a pre-constructed graph neural network for training, to obtain a near approach training result; Figure 2The principle schematic diagram of the near approach event prediction method provided by the embodiment of the present application is shown in the specific embodiment, as shown in Figure 1 and Figure 2 As shown, in order to obtain a target graph neural network with high precision that can be directly used for satellite near approach event prediction, a training data set needs to be constructed first. The training data set can be filtered and obtained from an open source satellite database.

[0021] At the same time, in order to improve the accuracy of the target graph neural network, in addition to training the initially constructed graph neural network through the training data set to obtain a near approach training result, a satellite knowledge graph is also constructed. It can be understood that the satellite knowledge graph can more accurately describe the relationship between satellites and satellites, and the relationship between satellites and orbits. Therefore, in an optional embodiment, the near approach event can be predicted based on the constructed satellite knowledge graph to obtain a near approach prediction result.

[0022] It should be noted that the pre-constructed graph neural network (Graph Neural Network, GNN for short) can be GCN (Graph Convolutional Network), GAT (Graph Attention Network), or GIN (Graph Isomorphism Network). The present application does not limit this.

[0023] In addition, it should also be noted that when predicting whether there is a near approach event between satellite nodes based on the satellite knowledge graph, a large language model in the satellite field can be used for reasoning and prediction, or a pre-trained prediction model can be used for prediction. The present application does not limit this.

[0024] S13: Construct a loss function by the near approach prediction result and the near approach training result; S14: Take minimizing the loss function as the goal, return to step S12 for iterative training to obtain a target graph neural network; S15: Input the feature data of the satellite to be predicted into the target graph neural network to obtain a target near approach prediction result of the satellite to be predicted.

[0025] As Figure 2As shown, in a specific embodiment, the training of the graph neural network is performed simultaneously to obtain the proximity training result, and the proximity event prediction is performed based on the satellite knowledge graph to obtain the proximity prediction result. For the same two satellite nodes, if the proximity training result is consistent with the proximity prediction result, it indicates that the current graph neural network has learned the proximity event fact between the two satellite nodes. If the proximity training result is inconsistent with the proximity prediction result, it indicates that the proximity knowledge learned by the graph neural network is incorrect, which will result in a low prediction accuracy of the target graph neural network for the proximity event.

[0026] Therefore, in an optional embodiment, as shown in Figure 2 The proximity training result and the proximity prediction result are combined to construct a loss function, which is used to represent the error between the proximity training result and the proximity prediction result.

[0027] In a specific embodiment, it can be understood that when the loss function is minimized, the error between the proximity training result and the proximity prediction result is minimized, and at this time, the prediction result of the graph neural network is closer to the true result. Therefore, as shown in Figure 2 After obtaining the loss function, it is determined whether the current loss function is the minimum value. If the current loss function does not reach the minimum value, it indicates that the error between the proximity training result and the proximity prediction result is not within an acceptable range, and the graph neural network needs to be continuously trained. At this time, step S12 is returned, that is, the training data set is continuously input to iteratively train the graph neural network. Of course, if it is the minimum value, it indicates that the graph neural network obtained by the current last iteration training meets the prediction accuracy requirement, and the graph neural network obtained by the current last iteration training can be directly used as the target graph neural network.

[0028] The target graph neural network trained by the above method can be directly applied to the prediction of satellite proximity events. Specifically, the satellite related information to be predicted is input into the target graph neural network, and thus the proximity event prediction result of the satellite to be predicted can be directly obtained, so that the satellite that may have a proximity event in the future can be intervened to avoid the occurrence of on-orbit collision.

[0029] Therefore, the proximity event prediction method provided by the embodiments of the present application converts the satellite and the orbit into a graph structure form of a knowledge graph, can quickly predict whether there is a proximity event between the satellite nodes according to the features of each node in the knowledge graph. At the same time, the graph neural network is trained by the training data set, so that the graph neural network processes complex nonlinear relationships to learn the proximity events between the satellites. Further, the loss function constructed by the proximity prediction result and the proximity training result is minimized to complete the training of the graph neural network, obtain a target graph neural network that can be directly used for proximity event prediction, avoid catastrophic on-orbit collisions caused by proximity events, and realize the sustainable development of the orbital environment.

[0030] In an optional embodiment, based on the satellite knowledge graph, it is predicted whether there is a close approach event between any two satellite nodes, and a close approach prediction result is obtained, including: From the satellite knowledge graph, node features of each entity node are extracted; the entity nodes include satellite nodes and orbit nodes; The node features are mapped to a unified vector space to obtain feature vectors; The feature vectors are normalized to obtain target feature vectors; The element-by-element multiplication result of any two target feature vectors is input into a multi-layer perception mechanism to obtain a target output result; According to the target output result, it is determined whether there is a close approach event between the two satellite nodes, and a close approach prediction result is obtained.

[0031] Figure 3 A principle diagram of node feature extraction provided by the embodiments of the present application, in specific embodiments, it can be understood that each entity node in the satellite knowledge graph has its own features, but these features may not be sufficient to express the semantic information of the node. For example, a satellite node may only have a satellite name as a feature, and the features of the entity nodes are extracted as much as possible to improve the accuracy of close approach event prediction, in an optional embodiment, as shown in Figure 3 , the positions of each entity node in the satellite knowledge graph and the relationships with other entity nodes can be learned through a graph neural network GNN, so as to obtain rich feature representations of each entity node, that is, the satellite node features and the orbit node features are extracted through the GNN.

[0032] In another optional embodiment, the satellite knowledge graph can also be inferred and analyzed by a large language model in the satellite field to extract node feature representations of each entity node. The method for improving node feature representation is not limited in the present application.

[0033] It should be noted that in specific embodiments, the entity nodes in the satellite knowledge graph include satellite nodes and orbit nodes, and when extracting node features, the node features of the two types of nodes are extracted because the two types of nodes are closely related. In order to obtain more rich node features.

[0034] Further, in order to facilitate subsequent prediction of close approach events based on node features, the node features need to be processed. First, the node features can be filtered and screened to eliminate outliers, and further, the node features are mapped to a unified vector space to obtain feature vectors, thereby ensuring that the node features are converted into learnable quantitative data.

[0035] In an optional embodiment, to ensure data stability and further improve the accuracy of proximity prediction results, the feature vectors are normalized to obtain a target feature vector. Then, any two target feature vectors are element-wise multiplied, and the multiplication result is input into the multi-layer perception mechanism to obtain the final target output result.

[0036] Specifically, in an optional embodiment, the calculation formula is as follows: (1) in, is the approximate prediction result, and its value range is . is the activation function, It is a multi-layer perception mechanism. is element-wise multiplication, and Respectively represent Satellite nodes and The target feature vector of the satellite nodes is the feature vector obtained due to the node features of two different satellite nodes.

[0037] In the above process of determining the target output result, it can be understood as first local crossover and then global fusion. That is, first multiply the two target feature vectors element by element, and then put them into a multi-layer perception mechanism. Learn nonlinear interactions.

[0038] In an optional embodiment, it is possible to determine whether there is a proximity event between two satellite nodes based on the target output result. Specifically, when the proximity prediction result When the proximity prediction result is When , it indicates that there is no proximity event between the two satellite nodes.

[0039] In an optional embodiment, node features of each entity node are extracted from the satellite knowledge graph, including: Constructing a vertex set in a hypergraph based on entity nodes, and constructing a hyperedge set in a hypergraph based on whether there is an edge between any two entity nodes; Constructing a first adjacency matrix based on the corresponding hyperedges between any two satellite nodes, and constructing a second adjacency matrix based on the corresponding hyperedges between the satellite nodes and the orbital nodes; Concatenate the first adjacency matrix and the second adjacency matrix to obtain the target adjacency matrix; Based on the target adjacency matrix, the features of each entity node are aggregated through the specified graph neural network to obtain the node features.

[0040] In a specific embodiment, a hypergraph is first constructed hypergraph includes a vertex set , a hyperedge set and a weight matrix . Wherein, the weight matrix is a diagonal matrix, and each hyperedge corresponds to a weight .

[0041] The vertex set is constructed based on the entity nodes in the satellite knowledge graph, and the hyperedge set is constructed based on whether there is an edge between any two entity nodes. Specifically, it includes whether there is an edge between satellite nodes and satellite nodes, and whether there is an edge between satellite nodes and orbit nodes.

[0042] The first adjacency matrix between satellite nodes and the second adjacency matrix between satellite nodes and orbit nodes can be constructed through the vertex set and the hyperedge set . Specifically, the first adjacency matrix can be constructed according to the corresponding hyperedge between any two satellite nodes, and the second adjacency matrix can be constructed according to the hyperedge between satellite nodes and orbit nodes.

[0043] Further, the first adjacency matrix and the second adjacency matrix are spliced to obtain the target adjacency matrix . Specifically, the splicing formula is formula (2): (2) Wherein, represents the feature splicing operation.

[0044] It can be understood that the target adjacency matrix is a matrix, and the element in the target adjacency matrix indicates whether the vertex (entity node) belongs to the hyperedge . In an optional embodiment, 1 can be used to represent belonging, and 0 can be used to represent not belonging.

[0045] In an optional embodiment, the node features can be aggregated by specifying a graph neural network, thereby obtaining rich node feature representations of each entity node. Specifically, see formula (3): (3) Wherein, is the The node characteristics of entity nodes, is a learnable linear transformation, is the aggregation function (e.g., attention mechanism, etc.), is the vertex degree matrix, is the target adjacency matrix, is the weight matrix, is the inverse of the edge degree matrix, For the The adjacent nodes of the entity node, that is, The neighbors of each vertex. (Entity Node ) belongs to the vertex set ,Right now, Each vertex Corresponding to a target feature vector ,in, , thus forming the input feature matrix .

[0046] In a specific embodiment, the present application does not limit the type of loss function constructed. In an optional embodiment, in order to ensure clearer gradients, more efficient training, and avoid gradient saturation, a binary cross-entropy loss function (Binary Cross-Entropy Loss, abbreviated as BCE Loss) can be selected.

[0047] Specifically, the calculation formula of BCE Loss is formula (4): (4) in, is the loss function value, is the training data set, For the Satellite nodes and The proximity training results of satellite nodes, The approximate prediction result.

[0048] In an optional embodiment, the training data set also includes a test data set, and the method for predicting proximity events includes: Input the test data set into the target graph neural network to obtain the target proximity prediction result; Based on the target proximity prediction results, determine the quality score used to evaluate the confidence of the target graph neural network proximity event prediction; When the quality score is not greater than the threshold, returning to the step of constructing a training dataset about satellite close approach events, and executing subsequent steps.

[0049] In an optional embodiment, as Figure 2As shown, in order to further improve the prediction accuracy of the close approach event, after obtaining the target graph neural network, the target graph neural network can be tested by using the test data set. Specifically, the test data set is input into the target graph neural network, and based on the target close approach prediction result, the quality score is calculated.

[0050] The quality score is used to evaluate the confidence of the target graph neural network in predicting close approach events, that is, to evaluate whether the accuracy of the target graph neural network meets the requirements. When the quality score is greater than the threshold value, it indicates that the accuracy of the target graph neural network meets the expectations, and the target graph neural network can be used as the final graph neural network, that is, as the final network for predicting satellite close approach events. When the quality score is not greater than the threshold value, the training data set is continuously input into the graph neural network for iterative training until a network that meets the conditions is obtained. Figure 2 As shown, the training data set is continuously input into the graph neural network for iterative training until a network that meets the conditions is obtained.

[0051] In an optional embodiment, the satellite knowledge graph is constructed, including: obtaining a target data set related to satellite close approach events; the satellites and orbits in the target data set are taken as entity nodes in the satellite knowledge graph; According to the close approach events between the satellite nodes and the running relationship between the satellite nodes and the orbit nodes, the edges between the entity nodes are generated; the satellite parameters are injected into the corresponding satellite nodes, and the orbit parameters are injected into the corresponding orbit nodes to construct the satellite knowledge graph.

[0052] In a specific embodiment, the target data set related to satellite close approach events is obtained from an open source satellite database. Further, the satellites and orbits in the target data set are taken as entity nodes, as shown in Figure 3 As shown, the satellite knowledge graph includes two types of nodes, satellite nodes and orbit nodes. In an optional embodiment, the satellite knowledge graph can include 27433 satellite nodes and 27433 orbit nodes.

[0053] The satellite knowledge graph includes two types of edges, as shown in Figure 3 One type is the edge between the satellite nodes, which is used to represent whether the two satellite nodes corresponding to the satellites have a close approach event at a certain time. In addition, the edge also includes the edge between the satellite node and the orbit node, which is used to represent the running of the satellite node corresponding to the satellite on the orbit node corresponding to the orbit. Therefore, in a specific embodiment, according to the close approach events between the satellite nodes and the running relationship between the satellite nodes and the orbit nodes, the edges between the entity nodes can be generated. In an optional embodiment, the satellite knowledge graph can include 71572 satellite close approach relationship edges and 27433 satellite running and orbit relationship edges.

[0054] A large amount of information about each satellite node and orbit node is also included in the target data set. In order to ensure the prediction accuracy of subsequent close approach events, satellite parameters are injected into the corresponding satellite node, and orbit parameters are injected into the corresponding orbit node. The satellite parameters at least include satellite number, payload and physical parameters, and the orbit parameters at least include orbit semi-major axis, orbit inclination and eccentricity.

[0055] In the above embodiment, the prediction method for the close approach event is described in detail, and the application also provides a corresponding embodiment of a prediction device for the close approach event.

[0056] Figure 4 A structural schematic diagram of a prediction device for a close approach event provided by an embodiment of the application is shown in Figure 4 The device comprises: A target construction module 40 is configured to construct a training data set for satellite close approach events and construct a satellite knowledge graph. A prediction module 41 is configured to predict whether there is a close approach event between any two satellite nodes based on the satellite knowledge graph, and obtain a close approach prediction result. A network training module 42 is configured to input the training data set into a pre-constructed graph neural network for training, and obtain a close approach training result. A loss function is constructed through the close approach prediction result and the close approach training result. The prediction module is returned to call with the goal of minimizing the loss function, and a target graph neural network is obtained. An input module 43 is configured to input feature data of a satellite to be predicted into the target graph neural network, and obtain a target close approach prediction result of the satellite to be predicted.

[0057] In addition, the prediction device for the close approach event provided by the embodiment of the application further comprises: A node feature extraction module is configured to extract node features of each entity node from the satellite knowledge graph. The entity nodes include satellite nodes and orbit nodes. A mapping module is configured to map the node features to a unified vector space to obtain feature vectors. A normalization module is configured to normalize the feature vectors to obtain target feature vectors. A multi-layer perception mechanism module is configured to input the element-by-element multiplication result of any two target feature vectors into a multi-layer perception mechanism to obtain a target output result. A close approach event determination module is configured to determine whether there is a close approach event between two satellite nodes according to the target output result, and obtain a close approach prediction result.

[0058] A hypergraph construction module is configured to construct a vertex set in a hypergraph based on the entity nodes, and construct a hyperedge set in the hypergraph based on whether there is an edge between any two entity nodes. An adjacency matrix construction module is configured to construct a first adjacency matrix according to the corresponding hyperedge between any two satellite nodes and construct a second adjacency matrix according to the corresponding hyperedge between the satellite node and the orbit node; A matrix splicing module is configured to splice the first adjacency matrix and the second adjacency matrix to obtain a target adjacency matrix. A feature aggregation module is configured to aggregate the features of each entity node based on the target adjacency matrix by using a specified graph neural network to obtain node features.

[0059] A test module is configured to input a test data set into the target graph neural network to obtain a target proximity prediction result. A quality score determination module is configured to determine a quality score for evaluating the proximity event prediction confidence of the target graph neural network according to the target proximity prediction result, and return to call the target construction module when the quality score is not greater than a threshold.

[0060] A target data set acquisition module is configured to acquire a target data set about satellite proximity events. An entity node determination module is configured to take the satellites and orbits in the target data set as entity nodes in the satellite knowledge graph. An edge generation module is configured to generate edges between the entity nodes according to the proximity events between the satellite nodes and the running relationship between the satellite nodes and the orbit nodes. An injection module is configured to inject satellite parameters into the corresponding satellite nodes and inject orbit parameters into the corresponding orbit nodes to construct the satellite knowledge graph.

[0061] Figure 5 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 1. Figure 5 As shown in FIG. 1, the electronic device includes a memory 50 configured to store a computer program. A processor 51 is configured to execute the computer program to implement the steps of the proximity event prediction method mentioned in the above embodiments.

[0062] The electronic device provided by the embodiment can include but is not limited to a notebook computer or a desktop computer, etc.

[0063] The processor 51 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 51 can be implemented in at least one of a hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic array (PLA), etc. The processor 51 can also include a main processor and a coprocessor. The main processor is a processor for processing data in a wake-up state, also referred to as a central processing unit (CPU). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 51 can be integrated with a graphics processor (GPU) for rendering and drawing content to be displayed by the display screen. In some embodiments, the processor 51 can further include an artificial intelligence (AI) processor for processing machine learning-related computing operations.

[0064] The memory 50 can include one or more computer-readable storage media that can be non-transitory. The memory 50 can further include a high-speed random access memory, and a nonvolatile memory such as one or more disk storage devices, flash storage devices. In this embodiment, the memory 50 is at least used to store the following computer program 501, wherein the computer program is loaded and executed by the processor 51, and can implement the related steps of the near-contact event prediction method disclosed in any of the preceding embodiments. In addition, the resources stored in the memory 50 can further include an operating system 502 and data 503, etc., and the storage mode can be temporary storage or permanent storage. The operating system 502 can include Windows, Unix, Linux, etc. The data 503 can include, but is not limited to, related data involved in the near-contact event prediction method, etc.

[0065] In some embodiments, the electronic device can further include a display screen 52, an input / output interface 53, a communication interface 54, a power supply 55, and a communication bus 56.

[0066] Those skilled in the art can understand that, Figure 5 The structure shown in the figure does not constitute a limitation on the electronic device, and can include more or fewer components than those shown.

[0067] The electronic device provided by the embodiment of the present application comprises a memory and a processor. The processor can realize the prediction method of the near event in the above embodiment when executing the program stored in the memory.

[0068] It should be noted that, although the operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring the operations to be performed in the particular order shown or sequentially, or requiring all of the illustrated operations to be performed to achieve a desired result. In some cases, multi-tasking and parallel processing can be advantageous. In addition, the separation of various system modules and components in the above embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

Claims

1. A method for predicting a near event, characterized in that: The method comprises: Build a training dataset of satellite close approach events and construct a satellite knowledge graph; Based on the satellite knowledge graph, predict whether there is a proximity event between any two satellite nodes, and obtain a proximity prediction result; Input the training data set into a pre-built graph neural network for training to obtain proximity training results; Constructing a loss function based on the proximity prediction result and the proximity training result; With the goal of minimizing the loss function, returning to the step of inputting the training dataset into the pre-built graph neural network to obtain a proximity training result, performing iterative training to obtain a target graph neural network; The characteristic data of the satellite to be predicted is input into the target graph neural network to obtain the target proximity prediction result of the satellite to be predicted.

2. The method for predicting a close event according to claim 1, wherein: The step of predicting whether there is a proximity event between any two satellite nodes based on the satellite knowledge graph to obtain a proximity prediction result includes: Extracting node features of each entity node from the satellite knowledge graph; the entity node includes the satellite node and the orbit node; Mapping the node features to a unified vector space to obtain a feature vector; Normalizing the feature vector to obtain a target feature vector; Input the result of element-wise multiplication of any two target feature vectors into the multi-layer perception mechanism to obtain the target output result; According to the target output result, it is determined whether there is a proximity event between the two satellite nodes to obtain the proximity prediction result.

3. The method for predicting a close event according to claim 2, wherein: Extracting node features of each entity node from the satellite knowledge graph includes: Constructing a vertex set in a hypergraph based on the entity nodes, and constructing a hyperedge set in the hypergraph based on whether there is an edge between any two entity nodes; Constructing a first adjacency matrix based on the corresponding hyperedges between any two of the satellite nodes, and constructing a second adjacency matrix based on the corresponding hyperedges between the satellite nodes and the orbital nodes; Concatenating the first adjacency matrix and the second adjacency matrix to obtain a target adjacency matrix; Based on the target adjacency matrix, the features of each of the entity nodes are aggregated through a specified graph neural network to obtain the node features.

4. The method for predicting a close event according to claim 1, wherein: The loss function is a binary cross entropy loss function.

5. The method for predicting a close event according to claim 1, wherein: The training data set also includes a test data set, and the method includes: Input the test data set into the target graph neural network to obtain a target proximity prediction result; Determining, based on the target proximity prediction result, a quality score for evaluating the confidence level of the target graph neural network proximity event prediction; When the quality score is not greater than the threshold, the method returns to the step of constructing a training data set for satellite close approach events and executes subsequent steps.

6. The method for predicting a near event according to claim 1, wherein: Build a satellite knowledge graph, including: Acquire targeted datasets on satellite close approach events; The satellites and orbits in the target dataset are used as entity nodes in the satellite knowledge graph; Generate edges between the entity nodes based on proximity events between satellite nodes and operational relationships between the satellite nodes and orbital nodes; Satellite parameters are injected into the corresponding satellite nodes, and orbit parameters are injected into the corresponding orbit nodes to construct the satellite knowledge graph.

7. The method for predicting a close event according to claim 6, wherein: The satellite parameters include at least satellite number, payload and physical parameters, and the orbit parameters include at least orbit semi-major axis, orbit inclination and eccentricity.

8. A device for predicting a near event, characterized in that: The device comprises: The target construction module is used to construct a training dataset on satellite close approach events and build a satellite knowledge graph; A prediction module, configured to predict whether there is a proximity event between any two satellite nodes based on the satellite knowledge graph, and obtain a proximity prediction result; A network training module is configured to input the training dataset into a pre-built graph neural network for training to obtain a proximity training result; construct a loss function based on the proximity prediction result and the proximity training result; and, with the goal of minimizing the loss function, return to the step of inputting the training dataset into the pre-built graph neural network to obtain the proximity training result, perform iterative training, and obtain a target graph neural network; The input module is used to input the characteristic data of the satellite to be predicted into the target graph neural network to obtain the target proximity prediction result of the satellite to be predicted.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the program, the steps of the method for predicting a proximity event according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for predicting a proximity event according to any one of claims 1 to 7 are implemented.

Images