Graph embedding-based satellite fault diagnosis method and apparatus, device, and storage medium

Abstract

Disclosed in the present invention are a graph embedding-based satellite fault diagnosis method and apparatus, a device, and a storage medium. The method comprises: obtaining a fault phenomenon query statement, and analyzing the fault phenomenon query statement to determine fault elements; constructing a virtual sub-graph on the basis of the fault elements, determining candidate sub-graphs on the basis of the fault elements in combination with a pre-constructed satellite fault knowledge graph, determining a first target vector corresponding to the virtual sub-graph and second target vectors corresponding to the candidate sub-graphs, and calculating the similarity between the first target vector and each second target vector to obtain the similarity between each candidate sub-graph and the virtual sub-graph; and performing fault diagnosis on the basis of the similarity corresponding to each candidate sub-graph in combination with the satellite fault knowledge graph to obtain a fault diagnosis result. The problems of inaccurate satellite fault diagnosis result, resource consumption, and low interpretability are solved; the fault diagnosis result is more accurate, computing power resources are saved, and the interpretability of fault causes is high.

Description

Satellite fault diagnosis method, device, equipment and storage medium based on graph embedding Technical Field

[0001] The present invention relates to the technical field of satellite fault diagnosis, and in particular to a satellite fault diagnosis method, apparatus, device and storage medium based on graph embedding. Background Art

[0002] In recent years, with the rapid development of aerospace technology, a wide range of satellite types have been launched worldwide, including communications, navigation, and remote sensing satellites. These satellites play a vital role in communications, navigation, meteorology, and other industries. Satellite failures can significantly impact people's livelihoods, disrupting essential services like communications and meteorology, and causing significant economic losses. Rapidly locating satellite faults and assisting with fault diagnosis has become a key research goal in the aerospace field, crucial for improving satellite reliability and timeliness.

[0003] During satellite in-orbit operations, numerous fault recovery reports are generated to record the troubleshooting process. Existing technologies typically include traditional rule-based approaches and deep learning-based approaches. Traditional rule-based approaches typically acquire real-time equipment operating data and then match it against expert-defined fault rules. The cause of the fault is determined based on different thresholds, or the cause is determined directly based on key operational indicators. Deep learning-based approaches require constructing features of the corresponding fault phenomena based on a knowledge graph, and then training a fault diagnosis model through feature engineering.

[0004] However, fault diagnosis based on traditional rules cannot adapt to fault information outside the rules because the rules are too limited. Key indicators cannot represent the cause of the entire fault, and the fault diagnosis results are not accurate enough. In deep learning-based solutions, training is required for each new fault report, which consumes a lot of computing power resources and requires a large amount of data to train a high-quality fault diagnosis model. The fault cause obtained by the model is not very explainable. Summary of the Invention

[0005] The present invention provides a satellite fault diagnosis method, apparatus, device and storage medium based on graph embedding to solve the problems of inaccurate satellite fault diagnosis results, waste of resources and poor interpretability.

[0006] According to one aspect of the present invention, a satellite fault diagnosis method based on graph embedding is provided, comprising:

[0007] Obtaining a fault phenomenon query statement, analyzing the fault phenomenon query statement, and determining the fault factor;

[0008] Constructing a virtual subgraph based on each of the fault elements, and determining a candidate subgraph based on each of the fault elements in combination with a pre-constructed satellite fault knowledge graph, determining a first target vector corresponding to the virtual subgraph and a second target vector corresponding to each of the candidate subgraphs, and calculating a similarity between the first target vector and each of the second target vectors to obtain a similarity between each of the candidate subgraphs and the virtual subgraph; wherein the first target vector and the second target vector are determined based on a graph embedding algorithm;

[0009] Fault diagnosis is performed based on the similarity corresponding to each of the candidate subgraphs and combined with the satellite fault knowledge graph to obtain a fault diagnosis result.

[0010] According to another aspect of the present invention, a satellite fault diagnosis device based on graph embedding is provided, comprising:

[0011] A query statement acquisition module is used to acquire a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine the fault factor;

[0012] a subgraph similarity calculation module, configured to construct a virtual subgraph based on each of the fault elements, determine a candidate subgraph based on each of the fault elements in combination with a pre-constructed satellite fault knowledge graph, determine a first target vector corresponding to the virtual subgraph and a second target vector corresponding to each of the candidate subgraphs, and calculate the similarity between the first target vector and each of the second target vectors to obtain a similarity between each of the candidate subgraphs and the virtual subgraph; wherein the first target vector and the second target vector are determined based on a graph embedding algorithm;

[0013] The fault diagnosis module is used to perform fault diagnosis based on the similarity corresponding to each candidate subgraph and the satellite fault knowledge graph to obtain a fault diagnosis result.

[0014] According to another aspect of the present invention, an electronic device is provided, comprising:

[0015] at least one processor, and a memory communicatively coupled to the at least one processor;

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the satellite fault diagnosis method based on graph embedding described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the satellite fault diagnosis method based on graph embedding according to any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention obtains a fault phenomenon query statement, analyzes the fault phenomenon query statement, and determines the fault factor; constructs a virtual subgraph and a candidate subgraph based on the fault factor to realize image search, and further determines the first target vector corresponding to the virtual subgraph and the second target vector corresponding to each candidate subgraph based on the graph embedding algorithm, realizes the conversion of nodes to the same spatial vector, and facilitates subsequent fault diagnosis; by calculating the similarity of the first target vector and each second target vector, the similarity between each candidate subgraph and the virtual subgraph is obtained, and then fault diagnosis is performed through the satellite fault knowledge graph, which solves the problems of inaccurate satellite fault diagnosis results, resource consumption and poor interpretability; the diagnosis process is implemented based on the satellite fault knowledge graph, and satellite fault diagnosis can be achieved without being limited to rules, and the fault diagnosis results are more accurate; and the satellite fault knowledge graph only needs to be constructed once, without the need for model training, saving computing power resources, and satellite fault diagnosis through the satellite fault knowledge graph has strong interpretability of the fault cause, which can help relevant personnel quickly locate the corresponding fault cause.

[0019] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] FIG1 is a flowchart of a satellite fault diagnosis method based on graph embedding according to a first embodiment of the present invention;

[0022] FIG2 is a flow chart of a satellite fault diagnosis method based on graph embedding according to a second embodiment of the present invention;

[0023] FIG3 is a diagram illustrating an implementation example of a candidate subgraph retrieval provided according to the second embodiment of the present invention;

[0024] FIG4 is a diagram illustrating an example of a process for constructing a satellite fault knowledge graph according to a second embodiment of the present invention;

[0025] FIG5 is a diagram illustrating an example structure of a satellite fault knowledge graph provided according to a second embodiment of the present invention;

[0026] FIG6 is a schematic process diagram of a graph embedding algorithm provided according to the second embodiment of the present invention;

[0027] FIG7 is an example diagram of a fault diagnosis tree provided according to the second embodiment of the present invention;

[0028] FIG8 is a diagram illustrating an overall implementation example of satellite fault diagnosis according to a second embodiment of the present invention;

[0029] FIG9 is a schematic structural diagram of a satellite fault diagnosis device based on graph embedding according to a third embodiment of the present invention;

[0030] FIG10 is a schematic structural diagram of an electronic device for implementing a satellite fault diagnosis method based on graph embedding according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described 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 making creative efforts should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] Example 1

[0034] Figure 1 is a flow chart of a satellite fault diagnosis method based on graph embedding according to a first embodiment of the present invention. This embodiment is applicable to accurately diagnosing satellite faults. The method can be performed by a satellite fault diagnosis device based on graph embedding. The device can be implemented in hardware and / or software and can be configured in an electronic device. As shown in Figure 1, the method includes:

[0035] S101: Obtain a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine the fault factor.

[0036] In this embodiment, the fault phenomenon query statement can be specifically understood as including a statement describing a satellite fault. The fault phenomenon query statement can be input by the user. For example, during satellite operation, the satellite's status, parameters, and other information are monitored. If an anomaly is detected, it can be considered that a fault has occurred during satellite operation. At this time, the satellite fault can be analyzed in real time. The staff can form a fault phenomenon query statement based on the fault phenomenon generated by the satellite. The fault element can be specifically understood as information describing the fault, such as the device name, parameters, parameter phenomenon, etc.

[0037] The satellite fault diagnosis method based on graph embedding provided in the embodiments of the present application can be deployed in a system or platform. The system or platform has an information receiving function and can provide a window or button for the user to input information. The user enters a fault phenomenon query statement, and the system or platform, as the execution device, receives the fault phenomenon query statement. The fault phenomenon query statement is parsed to extract element information from the fault phenomenon query statement. The extracted information is supplemented by using a professional dictionary as a word segmentation dictionary based on a word segmentation model. The two parts of information are then fused to obtain the fault factor.

[0038] For example, using the fault phenomenon query "north shunt shell temperature anomaly" as an example, the UIE information extraction model extracts the corresponding element information from the fault phenomenon query, including (north shunt, equipment), (anomaly, parameter phenomenon), etc. The LAC word segmentation model uses a professional dictionary as the word segmentation dictionary to supplement the extracted information. Finally, the two pieces of information are fused to obtain the fault elements. Using the UIE information extraction model, a high-quality extraction model can be trained using a small amount of text corpus.

[0039] S102: Construct a virtual subgraph based on each fault factor, determine a candidate subgraph based on each fault factor in combination with a pre-constructed satellite fault knowledge graph, determine a first target vector corresponding to the virtual subgraph and a second target vector corresponding to each candidate subgraph, and calculate the similarity between the first target vector and each second target vector to obtain a similarity between each candidate subgraph and the virtual subgraph; wherein the first target vector and the second target vector are determined based on a graph embedding algorithm.

[0040] In this embodiment, the virtual subgraph can be specifically understood as a structural diagram that describes relevant information about the occurrence of a fault, which is composed of fault elements. The virtual subgraph may or may not exist in the satellite fault knowledge graph. The candidate subgraph can be specifically understood as a structural diagram that may have certain similarities or correlations with the virtual subgraph, which is used to assist the virtual subgraph in fault diagnosis. The satellite fault knowledge graph can be specifically understood as a knowledge graph constructed based on relevant information about satellite faults. The satellite fault knowledge graph contains a large amount of relevant information about satellite faults and can be used to diagnose satellite faults. The first target vector can be specifically understood as a vector obtained after the virtual subgraph is vectorized, which is usually a multidimensional vector; the second target vector can be specifically understood as a vector obtained after the candidate subgraph is vectorized, which is usually also a multidimensional vector. The dimensions of the first target vector and the second target vector are the same.

[0041] Relevant information about satellite faults is obtained in advance, such as reports on satellite faults. Triples are generated by analyzing the relevant information, and a satellite fault knowledge graph is constructed based on the triples. When constructing a virtual subgraph based on each fault factor, the virtual subgraph can be constructed using all fault factors or by selecting important fault factors. For example, for parameters of the same type, the more important ones are selected to construct a virtual subgraph. The satellite fault knowledge graph is searched based on each fault factor to obtain different candidate subgraphs. When searching the satellite fault knowledge graph, the search can be based on all fault factors or selected based on a subset of fault factors to obtain different virtual subgraphs. Each node in the virtual subgraph is vectorized to obtain a corresponding first target vector. Alternatively, different nodes are vectorized in advance and the vectorization results are stored. During satellite fault diagnosis, the corresponding vector is directly queried and used as the first target vector. Similarly, the nodes in the candidate subgraph are vectorized to obtain a second target vector, or the pre-vectorized vectorization results are queried to obtain the second target vector. Node vectorization is achieved through a graph embedding algorithm. This involves pre-vectorizing and storing nodes using the graph embedding algorithm, or by vectorizing nodes using the graph embedding algorithm during fault diagnosis. This vectorization of nodes through the graph embedding algorithm transforms data into a vector space, facilitating subsequent data processing and calculation. The similarity between the first target vector and each second target vector is calculated using a similarity calculation method, resulting in the similarity between each candidate subgraph and the virtual subgraph.

[0042] For example, an embodiment of the present application provides a similarity calculation method. After the virtual subgraph and the candidate subgraph are respectively constructed with vector representations, the two vector representations are in the same vector space dimension, that is, the first target similarity and the second target similarity are in the same vector space dimension. Cosine similarity is used to calculate the similarity between the two subgraphs. Cosine similarity is a commonly used measurement method for comparing sentence similarity. Assuming that A and B represent the first target vector and the second target vector, respectively, the calculation method of cosine similarity is as follows:

[0043] Here, · represents the dot product operation of two vectors, and |||| represents the modulus of the vector.

[0044] S103: Perform fault diagnosis based on the similarity corresponding to each candidate subgraph and the satellite fault knowledge graph to obtain a fault diagnosis result.

[0045] In this embodiment, the fault diagnosis results may include fault causes, fault probabilities, etc., and may also include corresponding treatment measures. The fault diagnosis results may be output and displayed in the form of documents, text, etc., or may be displayed in the form of fault trees, etc., which is more intuitive.

[0046] Candidate subgraphs can be screened based on their corresponding similarities. For example, the candidate subgraph with the highest similarity is selected as the optimal subgraph. The satellite fault knowledge graph is then queried for this optimal subgraph to determine the corresponding fault report and analyze the cause of the fault. Alternatively, each candidate subgraph can be queried for its corresponding fault report, the cause of the fault analyzed, and the probability of each fault cause calculated based on similarity to form a fault tree. Satellite faults can be diagnosed using the satellite fault knowledge graph to obtain corresponding fault diagnosis results, which can assist personnel in troubleshooting and repairs.

[0047] An embodiment of the present invention provides a satellite fault diagnosis method based on graph embedding. The method obtains a fault phenomenon query statement, analyzes the fault phenomenon query statement, and determines the fault factor. A virtual subgraph and a candidate subgraph are constructed based on the fault factor to realize image search. A first target vector corresponding to the virtual subgraph and a second target vector corresponding to each candidate subgraph are further determined based on a graph embedding algorithm to realize the conversion of nodes to the same spatial vector, facilitating subsequent fault diagnosis. The similarity between each candidate subgraph and the virtual subgraph is calculated by calculating the similarity between the first target vector and each second target vector, and then the fault diagnosis is performed using a satellite fault knowledge graph, thereby solving the problems of inaccurate satellite fault diagnosis results, resource consumption, and poor interpretability. The diagnosis process is implemented based on the satellite fault knowledge graph, and satellite fault diagnosis can be achieved without being limited to rules, resulting in more accurate fault diagnosis results. In addition, the satellite fault knowledge graph only needs to be constructed once, without the need for model training, saving computing power resources. Satellite fault diagnosis using the satellite fault knowledge graph has strong interpretability of fault causes, which can help relevant personnel quickly locate the corresponding fault causes.

[0048] Example 2

[0049] FIG2 is a flow chart of a satellite fault diagnosis method based on graph embedding provided by the second embodiment of the present invention. This embodiment is a refinement of the above embodiment. As shown in FIG2 , the method includes:

[0050] S201: Obtain a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine the fault factor.

[0051] S202: Take each fault element as a node of a virtual subgraph, and connect each node in the virtual subgraph to obtain a virtual subgraph.

[0052] All fault elements are regarded as a node of a virtual subgraph, and multiple nodes of the virtual subgraph are obtained. The nodes are connected, and the connection method can be two-by-two connection. The obtained structural diagram is the virtual subgraph.

[0053] S203 , querying a vector database to determine a first vector corresponding to each node in the virtual subgraph, where the vector database is constructed based on a graph embedding algorithm.

[0054] A vector database is pre-built based on a graph embedding algorithm. Nodes are vectorized using the graph embedding algorithm to obtain a vector corresponding to each node, which is then stored in the vector database. The vector database stores vectors corresponding to different nodes. After the virtual subgraph is constructed, the vector database is queried for each node in the virtual subgraph. For each node, the matching node in the vector database is determined, and the vector corresponding to the matching node is used as the first vector. When querying the vector database to determine matching nodes, nodes that are completely identical to nodes in the virtual subgraph can be identified as matching nodes. Alternatively, if no identical nodes exist, similar nodes can be identified as matching nodes, and so on.

[0055] S204: Determine a first target vector based on each first vector.

[0056] After obtaining the first vector corresponding to each node, each first vector is processed to obtain a first target vector. The first target vector can be determined based on each first vector by weighted summation or averaging of the first vectors. For example, if the first vector is a d-dimensional vector, each dimension of each first vector can be averaged to obtain the first target vector.

[0057] For example, the corresponding fault elements (north shunt, equipment), (shell temperature, parameter), and (abnormality, parameter phenomenon) in the fault phenomenon query statement are v1, v2, and v3 in the vector library, that is, the first vectors are v1, v2, and v3 respectively; the "virtual subgraph" vector calculation of the corresponding query statement is represented by the average value of the above vectors: Among them, V query is the first target vector, k is the total number of fault elements, v i is the i-th first vector.

[0058] S205: Determine the number of at least one subgraph node.

[0059] In this embodiment, the number of subgraph nodes can be specifically understood as the number of nodes in the candidate subgraph when constructing the candidate subgraph, that is, the number of fault factors selected to construct the candidate subgraph. The number of subgraph nodes can be pre-set and can be set to a fixed value, for example, 1, 2, or 3. It can also be set based on the number of fault factors. When the number of fault factors is large, the number of subgraph nodes can be relatively large, and when the number of fault factors is small, the number of subgraph nodes can be relatively small. The number of subgraph nodes can also be set to a fixed value, for example, if the number of subgraph nodes is 3, the number of subgraph nodes is 1, 2, and 3, respectively, for a total of 3. Or it can be set based on the number of fault factors. If the number of fault factors is 7, the number of subgraph nodes is 4, the number of subgraph nodes is 1, 2, 3, and 4, respectively, for a total of 4. If the number of fault factors is 8, the number of subgraph nodes is 5, the number of subgraph nodes is 1, 2, 3, 4, and 5, respectively, for a total of 5, and so on.

[0060] S206. For each subgraph node number, extract each fault element according to the subgraph node number to obtain at least one candidate element set, and query the satellite fault knowledge graph based on the candidate fault elements in each candidate element set to obtain a corresponding candidate subgraph.

[0061] In this embodiment, the candidate fault elements may be specifically understood as the fault elements used to determine the candidate subgraphs; and the candidate element set may be specifically understood as the set consisting of the candidate fault elements.

[0062] For each subgraph node, the same number of candidate fault factors is extracted from each fault factor according to the number of subgraph nodes to form a candidate factor set. The number of candidate factor sets corresponding to the same number of subgraph nodes can be one or more, depending on the total number of fault factors. When extracting candidate factor sets, all combinations of fault factors can be extracted, or only some combinations can be extracted. This can be achieved by setting a rule for the number of candidate factor sets. For each candidate factor set, the satellite fault knowledge graph is queried to obtain the candidate subgraph corresponding to this candidate factor set. The number of candidate subgraphs obtained can be one or more. The candidate subgraphs obtained from each candidate factor set constitute a candidate subgraph set.

[0063] S207 , querying a vector database to determine a second vector corresponding to each node in the candidate subgraph, where the vector database is constructed based on a graph embedding algorithm.

[0064] A vector database is pre-built based on a graph embedding algorithm, and different nodes and their corresponding vectors are stored in the vector database. The vector database is then queried for each node in the candidate subgraph. For each node, the matching node in the vector database is determined, and the vector corresponding to the matching node is used as the second vector. When querying the vector database to determine matching nodes, nodes that are exactly identical to nodes in the candidate subgraph can be identified as matching nodes. Alternatively, similar nodes can be identified as matching nodes if no identical nodes exist, and so on.

[0065] S208 : Determine a weight according to the number of subgraph nodes corresponding to the candidate subgraph.

[0066] Since candidate subgraphs are generated by querying the satellite fault knowledge graph to form a candidate element set based on the number of subgraph nodes, the corresponding relationship between the candidate subgraphs and the number of subgraph nodes can be determined simultaneously with the candidate subgraphs. Weights are pre-set based on the number of subgraph nodes; for example, the greater the number of subgraph nodes, the greater the weight. After the candidate subgraphs are determined, the corresponding weights are determined based on the pre-set relationship between the number of subgraph nodes and the weight, as well as the relationship between the candidate subgraphs and the number of subgraph nodes.

[0067] S209: Perform a weighted operation on each second vector based on the weight to obtain a second target vector.

[0068] A weighted operation is performed on each second vector using a weight, and a weighted sum of multiple second vectors is obtained to obtain a second target vector.

[0069] For example, the fault phenomenon query statement is "0A star north shunt shell temperature is abnormal." The fault factor entities obtained by the UIE extraction model and LAC word segmentation model are (0A star, star), (north shunt, equipment), (shell temperature, parameter), (abnormal, parameter phenomenon). Since there is a one-to-one correspondence between star and equipment in the knowledge model, the above fault factors are a combination of equipment, parameter, and parameter phenomenon to query the subgraph. The specific query process is as follows:

[0070] To ensure recall of candidate subgraphs, the query method is based on combinations of extracted fault elements. For example, G1 is used to represent the candidate element set, where l∈L, and L is the number of corresponding extracted elements. For example, if L is set to 3, G3 represents the combination of (device, parameter, parameter phenomenon), G2 represents the pairwise combinations of elements (device, parameter), (device, parameter phenomenon), and (parameter, parameter phenomenon), and G1 represents the single element combination (device), (parameter), and (parameter phenomenon).

[0071] The candidate element set G3 queries the corresponding candidate subgraph set in the satellite fault knowledge graph as follows: Set the weight W3>0, and query the corresponding fault phenomenon set of the candidate element set G2 in the satellite fault knowledge graph as follows: Set the weight W2>0, and query the corresponding fault phenomenon set of the candidate element set G1 in the satellite fault knowledge graph as follows: The weights are set to W1 > 0, and W3 > W2 > W1. This is because, based on subgraph structural considerations, the more fault elements a fault contains, the more structurally similar the fault phenomena are. Note that the candidate subgraph sets F3, F2, and F1 mentioned above may not all satisfy the query, so the corresponding candidate subgraph sets can be empty. All subgraphs included in candidate subgraph sets F3, F2, and F1 are candidate subgraphs.

[0072] According to the candidate subgraphs obtained from the above query, the vectors of the nodes corresponding to the candidate subgraphs in each candidate subgraph set are queried in the fault node vector library, namely: S is the query function of the vector library, which aims to convert the corresponding fault phenomenon into a vector q of the same dimension d. Therefore, the vector corresponding to each candidate subgraph in the above candidate subgraph set is represented by Q l :W1*q i , where l is the number of corresponding subgraph nodes, q i is the vector corresponding to each fault phenomenon (i.e. node).

[0073] In the embodiment of the present application, determining the first target vector and the second target vector can be performed in parallel or sequentially, without a strict order. FIG2 takes parallel execution as an example.

[0074] S210 : Calculate the similarity between the first target vector and each second target vector to obtain the similarity between each candidate subgraph and the virtual subgraph.

[0075] For example, Figure 3 provides an example implementation of candidate subgraph retrieval: A fault phenomenon query statement 31 is "north shunt shell temperature abnormal." The query statement is analyzed using the UIE information extraction model 32. The query statement is then analyzed using the LAC word segmentation model 33 in conjunction with a professional dictionary 34. Information fusion is then performed to obtain fault elements. A virtual subgraph 35 is constructed based on the fault elements. Based on the fault elements, the satellite fault knowledge graph 36 is queried to generate candidate subgraphs 37. Similarity is calculated between the virtual subgraph 35 and the candidate subgraphs 37. The optimal subgraph can also be selected based on the similarity.

[0076] As an optional embodiment of this embodiment, this optional embodiment further optimizes the construction of the satellite fault knowledge graph, including steps A1-A5:

[0077] A1. Obtain satellite zeroing report.

[0078] In this embodiment, a satellite return-to-zero report can be specifically understood as a file recording information related to satellite faults. During satellite in-orbit operation, numerous satellite return-to-zero reports are generated to record the satellite fault resolution process. Satellite return-to-zero reports are generated in advance by recording the satellite's operational process and can be stored in a cloud, memory, or other storage space. When constructing the satellite fault knowledge graph, satellite return-to-zero reports are read based on the storage address.

[0079] A2. Split the title of the satellite zeroing report to obtain at least one title.

[0080] The satellite zeroing report is split according to the title, and the corresponding report content is structured according to the document parsing algorithm. The title can be the problem name, problem overview, problem description, problem item, problem sub-item, problem location, event cause, detection result, mechanism analysis, disposal measures or next step.

[0081] The title type can be structured or unstructured, that is, the content corresponding to the title can be structured data or unstructured data. For each title, we determine whether it is structured or unstructured and process it using A3 or A4 to obtain the corresponding triple.

[0082] A3. If the title type is unstructured, extract the content corresponding to the title and construct triples based on the extracted knowledge content.

[0083] If the title type is unstructured, the content corresponding to the title is input into the information extraction model. The information extraction model extracts information from the content corresponding to the title, for example, the star, subsystem, equipment, parameter, and parameter phenomenon corresponding to the fault phenomenon (collectively referred to as the five elements). The extracted knowledge content is then combined into corresponding triples.

[0084] A4. If the title type is structured, construct a triple based on the content corresponding to the title.

[0085] If the title type is structured, the content corresponding to the title has certain rules and can be directly split to obtain the required content to construct triples.

[0086] A5. Perform knowledge fusion based on each triple to form a satellite fault knowledge graph.

[0087] For example, among the aforementioned titles, only the question item is unstructured, while the remaining titles are structured. When constructing a satellite fault knowledge graph, we can directly determine whether the title is a question item. If so, we input the question item's content into the information extraction model to extract the five elements. The extracted knowledge content is then combined into corresponding triples. If it is not a question item, the remaining content is combined with the fault report to form the corresponding fault report triple.

[0088] For example, FIG4 provides an example diagram of a process for constructing a satellite fault knowledge graph, including the following steps:

[0089] S401. Obtain a satellite zeroing report.

[0090] S402: Split the title of the satellite zeroing report.

[0091] S403. Determine whether each title is a question item in turn. If so, execute S404; otherwise, execute S406.

[0092] S404: Extract information.

[0093] S405: Extract the fault triples and execute S408.

[0094] Exemplarily, a fault triplet may be (fault phenomenon, fault phenomenon_parameter, parameter), (fault phenomenon, fault phenomenon_device, device), (star, star_subsystem, subsystem), and so on.

[0095] S406: Perform structured splitting.

[0096] S407: Split into fault report triplets.

[0097] Exemplarily, a fault report triplet may be (fault report, fault report_bottom event, bottom event), (fault report, fault report_fault classification, fault classification), (fault report, report corresponding document, report document), and the like.

[0098] S408, knowledge integration.

[0099] Exemplarily, the knowledge fusion may obtain (fault phenomenon, fault report_fault phenomenon, fault report).

[0100] S409: Form a satellite fault knowledge graph.

[0101] For example, FIG5 provides a structural example diagram of a satellite fault knowledge graph, which only exemplarily shows nodes in the satellite fault knowledge graph such as fault reports and fault phenomena.

[0102] As an optional embodiment of this embodiment, this optional embodiment is further optimized to include: vectorizing the nodes in the satellite fault knowledge graph based on a graph embedding algorithm, obtaining each node and its corresponding vector, and storing them in a vector database accordingly.

[0103] After completing the construction of the satellite fault knowledge graph, it is necessary to capture the structure and homogeneity of the five-element subgraphs that constitute the fault phenomenon to form a high-quality graph feature vector. This embodiment of the application uses a graph embedding algorithm to perform a conditional walk sampling sequence on the nodes in the satellite fault knowledge graph and then vectorizes them. This generates vectors of the same dimension for the nodes in the satellite fault knowledge graph and stores the nodes and their corresponding vectors in a vector database.

[0104] For example, let's take the graph embedding algorithm Node2vec as an example. Node2vec refers to node vectorization. This model is an improved version of the DeepWalk algorithm and uses the Skip-Gram model to learn node representations. Figure 6 provides a schematic diagram of the graph embedding algorithm.

[0105] The following describes how to train the fault phenomenon nodes and the five-element subgraph in the satellite fault knowledge graph to obtain a graph embedding model with the characteristics of composite satellite structural elements. First, the model mainly controls the propensity of the walk by setting two walk parameters, p and q, as shown in the figure above. Then, the walk is sampled according to the set bias. Finally, the sampled sequence is trained on word embedding. The algorithm principle can be divided into the following three steps:

[0106] (1) Randomly walk the nodes on the satellite fault knowledge graph to generate a set of node sequences.

[0107] (2) During the random walk, some probability distributions are used to determine the next node to be visited, thereby ensuring the diversity of the random walk.

[0108] (3) Take the node sequence as input and use the Skip-Gram model to learn the node representation.

[0109] Node2vec extends the Skip-Gram model in NLP tasks to network learning, with the goal of optimizing the following objective function:

[0110] In the above formula, V represents the node set in the network, and f is the node to feature representation R in the corresponding node set V. d The mapping function, d is the dimension of feature representation, P(N s (u)|f(u) represents the logarithmic probability of observing its neighborhood by the feature representation of a node, N s(u) is a node sequence generated by sampling the nodes in the neighborhood through a sampling strategy. The corresponding node sequence is obtained through random walk sampling. During the random walk, the node needs to select the next node to walk to based on the probability distribution, that is, to select from the neighbor nodes and second-degree nodes of the current node, which corresponds to the p and q parameters in Figure 4.

[0111] In the extended Skip-Gram learning process, each node is represented as a vector of the same dimension. The basic goal is to learn the representation vector of each node so that this vector can predict the neighboring nodes around a specific node. The node representation vector is trained by maximizing the conditional probability of the neighboring nodes. In order to ensure the symmetry of the feature space, the likelihood of each source node and neighboring node pair is modeled as a softmax unit as follows:

[0112] Among them, V is the entire node set, v j ∈N s (u) is the sampling sequence corresponding to the current node, and represents the vector dot product. The above formula predicts the probability of the representation vector corresponding to the current node's neighborhood given the representation vector of node u, assuming that the sub-neighborhood probabilities are independent of each other. By minimizing the relevant loss function, the node representation vector can be adjusted using a gradient descent algorithm to maximize Formula 1. After the above training process is completed, the feature vector matrix generated during the training process represents the vector features of the corresponding node. The graph embedding vector with satellite structural elements obtained through training is stored in the vector database.

[0113] S211. Perform fault diagnosis based on the similarity corresponding to each candidate subgraph and the satellite fault knowledge graph to obtain a fault diagnosis result.

[0114] During the fault diagnosis phase, after completing sub-graph matching, the fault cause probability model can be used to infer possible fault causes, their probabilities, and recommended treatment measures. This helps staff locate the fault point in a timely manner after discovering the fault phenomenon and use the recommended treatment measures as a guide. Combined with professional judgment, the fault problem can be quickly resolved and the satellite can resume normal operation.

[0115] Optionally, the fault diagnosis result includes a fault diagnosis tree and a fault report;

[0116] As an optional embodiment of this embodiment, this optional embodiment further performs fault diagnosis based on the similarity corresponding to each candidate subgraph in combination with the satellite fault knowledge graph, and obtains the fault diagnosis result optimized as B1-B2:

[0117] B1. Sort each candidate subgraph by similarity, query the satellite fault knowledge graph to determine the bottom event in the fault report corresponding to the fault phenomenon of the candidate subgraph, and calculate the probability value of the bottom event.

[0118] In the satellite fault tree analysis model, fault causes are typically presented as bottom events. Candidate subgraphs are first sorted by similarity to facilitate quick identification of the most proximate fault cause. The satellite fault knowledge graph is then queried to determine the fault phenomenon corresponding to the candidate subgraph. The fault report is then determined based on the relationship between the fault phenomenon and the fault report, and the bottom events within the fault report are then identified. The probability of a bottom event causing the corresponding fault phenomenon is determined by the ratio of the number of phenomena corresponding to that bottom event to the number of bottom events generated by that phenomenon. This analysis determines the probability of each bottom event occurring.

[0119] The bottom events obtained by the query in the embodiment of the present application can constitute a bottom event set. Since the probability of a corresponding fault phenomenon caused by a corresponding bottom event in the bottom event set depends on the ratio of the number of phenomena corresponding to the bottom event and the number of bottom events generated by the phenomenon, it conforms to the calculation principle of Bayesian conditional probability. The relationship between the fault phenomenon and the bottom event that may cause the fault event can be displayed using a Bayesian probability graph.

[0120] The principle of fault tree and probability of occurrence of bottom events is explained below:

[0121] Assume there are N random vectors X={x1,x2,...,x i ,...,x n} and the corresponding parent node where x i With the parent node P a (x i ), and assuming that under the parent node, the variables are independent of each other, that is:

[0122] x i Its parent node P a (x i ), the complete likelihood function is calculated as: P(x1,x2,...,x i ,...,x n )=∏P(x i |P a (x i )) (4)

[0123] Among them, P(x i |P a (x i )) represents a given parent node P a (x i), the random variable x i The conditional probability of taking a certain value. By learning from sample data, we can calculate the conditional probability distribution of each variable, and thus the joint probability distribution. Ultimately, through iterative calculations, we can obtain a complete fault tree, where the nodes on the tree represent the probability of the underlying event occurring.

[0124] B2. Form a fault diagnosis tree based on the probability value of each bottom event, and determine the fault report corresponding to each bottom event.

[0125] Each bottom event is formed into a fault diagnosis tree according to the hierarchical relationship and probability value, and the fault report corresponding to each bottom event is determined. The fault diagnosis tree and fault report are fed back to the user as the fault diagnosis result to help the user solve the fault problem.

[0126] Exemplarily, Figure 7 provides an example diagram of a fault diagnosis tree, which exemplarily shows the base events corresponding to different similar fault phenomena, as well as the probabilities corresponding to the base events, wherein similar fault phenomena can be determined based on the similarity between the virtual subgraph and the candidate subgraph, and p1, p2, p3, p11, p21-p25 in the figure are the corresponding probabilities.

[0127] For example, Figure 8 provides a diagram illustrating the overall implementation of satellite fault diagnosis. Satellite fault diagnosis is generally divided into four phases: preprocessing, graph embedding, subgraph retrieval, and fault diagnosis. In the preprocessing phase, satellite fault reports are analyzed and processed. A text extraction module extracts relevant fault knowledge and constructs a satellite fault knowledge graph. The text extraction module includes a UIE extraction model and a LAC word segmentation model. In the graph embedding phase, the entire satellite fault knowledge graph is embedded using a composite satellite element graph embedding algorithm. Corresponding nodes are converted to the same spatial vector and stored in a vector database. In the subgraph retrieval phase, the text extraction module extracts relevant entities from the fault phenomenon query. A subgraph construction algorithm is then used to construct a virtual subgraph and corresponding candidate subgraphs. Finally, a similarity algorithm is used to calculate and rank the similarity scores between the virtual subgraph and the candidate subgraphs. In the fault diagnosis phase, a fault diagnosis tree and corresponding scores for the corresponding bottom events in the fault report corresponding to the fault phenomenon are calculated based on the bottom events in the relevant candidate subgraphs retrieved in the subgraph retrieval phase.

[0128] The present invention provides a satellite fault diagnosis method based on graph embedding, addressing the issues of inaccurate satellite fault diagnosis results, resource consumption, and poor interpretability. By introducing an unsupervised graph embedding algorithm and combining it with a satellite fault knowledge graph, the method can quickly capture the structural characteristics of the corresponding fault nodes in the graph. Based on different diagnostic tasks, the model can be adjusted based on parameters to favor homogeneity and structure, eliminating the need for tedious feature engineering. The trained graph embedding model possesses the characteristics of composite satellite structural elements. By combining the structured knowledge of underlying events in the report, a fault tree for the corresponding fault can be dynamically generated, helping personnel quickly locate the cause of the fault.

[0129] Example 3

[0130] FIG9 is a schematic diagram of the structure of a satellite fault diagnosis device based on graph embedding according to the third embodiment of the present invention. As shown in FIG9 , the device includes: a query statement acquisition module 51 , a subgraph similarity calculation module 52 , and a fault diagnosis module 53 .

[0131] The query statement acquisition module 51 is used to acquire a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine the fault factor;

[0132] a subgraph similarity calculation module 52 for constructing a virtual subgraph and a candidate subgraph based on each of the fault elements, determining a first target vector corresponding to the virtual subgraph and a second target vector corresponding to each of the candidate subgraphs, and calculating the similarity between the first target vector and each of the second target vectors to obtain the similarity of each of the candidate subgraphs; wherein the first target vector and the second target vector are determined based on a graph embedding algorithm;

[0133] The fault diagnosis module 53 is used to perform fault diagnosis based on the similarity of each candidate subgraph in combination with a pre-built satellite fault knowledge graph to obtain a fault diagnosis result.

[0134] An embodiment of the present invention provides a satellite fault diagnosis method based on graph embedding. The method obtains a fault phenomenon query statement, analyzes the fault phenomenon query statement, and determines the fault factor. A virtual subgraph and a candidate subgraph are constructed based on the fault factor to realize image search. A first target vector corresponding to the virtual subgraph and a second target vector corresponding to each candidate subgraph are further determined based on a graph embedding algorithm to realize the conversion of nodes to the same spatial vector, facilitating subsequent fault diagnosis. The similarity between each candidate subgraph and the virtual subgraph is calculated by calculating the similarity between the first target vector and each second target vector, and then the fault diagnosis is performed using a satellite fault knowledge graph, thereby solving the problems of inaccurate satellite fault diagnosis results, resource consumption, and poor interpretability. The diagnosis process is implemented based on the satellite fault knowledge graph, and satellite fault diagnosis can be achieved without being limited to rules, resulting in more accurate fault diagnosis results. In addition, the satellite fault knowledge graph only needs to be constructed once, without the need for model training, saving computing power resources. Satellite fault diagnosis using the satellite fault knowledge graph has strong interpretability of fault causes, which can help relevant personnel quickly locate the corresponding fault causes.

[0135] Optionally, the subgraph similarity calculation module 52 includes:

[0136] The virtual subgraph determining unit is configured to use each of the fault elements as a node of a virtual subgraph, and connect each node in the virtual subgraph to obtain a virtual subgraph.

[0137] Optionally, the subgraph similarity calculation module 52 includes:

[0138] a node quantity determination unit, configured to determine the quantity of at least one subgraph node;

[0139] The candidate subgraph determining unit is configured to extract each of the fault elements according to the number of subgraph nodes to obtain at least one candidate element set, and query the satellite fault knowledge graph based on the candidate fault elements in each candidate element set to obtain a corresponding candidate subgraph.

[0140] Optionally, the subgraph similarity calculation module 52 includes:

[0141] a first vector determining unit, configured to query a vector database to determine a first vector corresponding to each node in the virtual subgraph, wherein the vector database is constructed based on a graph embedding algorithm;

[0142] The first target vector determining unit is configured to determine a first target vector based on each of the first vectors.

[0143] Optionally, the subgraph similarity calculation module 52 includes:

[0144] a second vector determining unit, configured to query a vector database to determine a second vector corresponding to each node in the candidate subgraph, wherein the vector database is constructed based on a graph embedding algorithm;

[0145] A weight determination unit, configured to determine a weight according to the number of subgraph nodes corresponding to the candidate subgraph;

[0146] The second target vector determining unit is configured to perform a weighted operation on each of the second vectors based on the weights to obtain a second target vector.

[0147] Optionally, the device further includes:

[0148] Satellite report acquisition module, used to obtain satellite zeroing reports;

[0149] A title splitting module is used to split the title of the satellite zeroing report to obtain at least one title;

[0150] A first construction module is configured to extract information from the content corresponding to the title if the title type is unstructured, and construct triples based on the extracted knowledge content;

[0151] A second construction module is configured to construct a triple according to the content corresponding to the title if the type of the title is structured;

[0152] The graph generation module is used to perform knowledge fusion based on each of the triples to form a satellite fault knowledge graph.

[0153] Optionally, the device further includes:

[0154] The vectorization module is used to vectorize the nodes in the satellite fault knowledge graph based on the graph embedding algorithm, obtain each node and its corresponding vector, and store them in the vector database accordingly.

[0155] Optionally, the fault diagnosis result includes a fault diagnosis tree and a fault report;

[0156] Optionally, the fault diagnosis module 53 includes:

[0157] A bottom event probability calculation module is used to sort the candidate subgraphs according to their similarity, query the satellite fault knowledge graph to determine the bottom events in the fault reports corresponding to the fault phenomena of the candidate subgraphs, and calculate the probability values ​​of the bottom events;

[0158] The fault diagnosis tree determination module is used to form a fault diagnosis tree based on the probability value of each of the bottom events and determine the fault report corresponding to each bottom event.

[0159] The satellite fault diagnosis device based on graph embedding provided by the embodiment of the present invention can execute the satellite fault diagnosis method based on graph embedding provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0160] Example 4

[0161] FIG10 shows a block diagram of an electronic device 60 that can be used to implement an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0162] As shown in FIG10 , the electronic device 60 includes at least one processor 61 and a memory, such as a read-only memory (ROM) 62 and a random access memory (RAM) 63, that is communicatively connected to the at least one processor 61. The memory stores a computer program that can be executed by the at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 into the random access memory (RAM) 63. Various programs and data required for the operation of the electronic device 60 can also be stored in the RAM 63. The processor 61, ROM 62, and RAM 63 are connected to each other via a bus 64. An input / output (I / O) interface 65 is also connected to the bus 64.

[0163] Multiple components in the electronic device 60 are connected to the I / O interface 65, including an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a magnetic disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0164] Processor 61 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. Processor 61 executes the various methods and processes described above, such as the satellite fault diagnosis method based on graph embedding.

[0165] In some embodiments, the satellite fault diagnosis method based on graph embedding can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as storage unit 68. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 60 via ROM 62 and / or communication unit 69. When the computer program is loaded into RAM 63 and executed by processor 61, one or more steps of the satellite fault diagnosis method based on graph embedding described above can be performed. Alternatively, in other embodiments, processor 61 can be configured to execute the satellite fault diagnosis method based on graph embedding by any other appropriate means (e.g., by means of firmware).

[0166] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0167] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0168] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0169] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0170] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0171] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0172] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0173] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A satellite fault diagnosis method based on graph embedding, characterized in that Including: Obtain a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine fault elements; Construct a virtual subgraph based on each of the fault elements, and based on each of the fault elements, combine with a pre-constructed satellite fault knowledge graph to determine candidate subgraphs, determine a first target vector corresponding to the virtual subgraph and second target vectors corresponding to each of the candidate subgraphs, and calculate the similarity between the first target vector and each of the second target vectors to obtain the similarity between each of the candidate subgraphs and the virtual subgraph; wherein, the first target vector and the second target vector are determined based on a graph embedding algorithm; Perform fault diagnosis based on the similarity corresponding to each of the candidate subgraphs in combination with the satellite fault knowledge graph to obtain a fault diagnosis result.

2. The method according to claim 1, wherein The constructing a virtual subgraph based on each of the fault elements includes: Use each of the fault elements as nodes of the virtual subgraph, and connect each of the nodes in the virtual subgraph to obtain a virtual subgraph.

3. The method according to claim 1, characterized in that, The determining candidate subgraphs based on each of the fault elements in combination with a pre-constructed satellite fault knowledge graph includes: Determine at least one subgraph node number; For each subgraph node number, extract each of the fault elements according to the subgraph node number to obtain at least one candidate element set, and query the satellite fault knowledge graph based on the candidate fault elements in each of the candidate element sets to obtain corresponding candidate subgraphs.

4. The method according to claim 1, wherein The determining a first target vector corresponding to the virtual subgraph includes: Query a vector database to determine a first vector corresponding to each node in the virtual subgraph, and the vector database is constructed based on a graph embedding algorithm; Determine a first target vector based on each of the first vectors.

5. The method according to claim 3, characterized in that, Determine a second target vector corresponding to the candidate subgraph, including: Query a vector database to determine a second vector corresponding to each node in the candidate subgraph, and the vector database is constructed based on a graph embedding algorithm; Determine a weight according to the subgraph node number corresponding to the candidate subgraph; Perform a weighted operation on each of the second vectors based on the weight to obtain a second target vector.

6. The method according to any one of claims 1-5, characterized in that, The steps of constructing the satellite fault knowledge graph include: Obtain a satellite problem-solving report; Perform title splitting on the satellite problem-solving report to obtain at least one title; If the type of the title is unstructured, perform information extraction on the content corresponding to the title, and construct triples according to the extracted knowledge content; If the type of the title is structured, construct triples according to the content corresponding to the title; Perform knowledge fusion based on each of the triples to form a satellite fault knowledge graph.

7. The method according to claim 6, characterized in that, It also includes: Vectorize the nodes in the satellite fault knowledge graph based on a graph embedding algorithm to obtain each of the nodes and their corresponding vectors, and store them in the vector database correspondingly.

8. The method according to claim 1, wherein The fault diagnosis result includes a fault diagnosis tree and a fault report; the performing fault diagnosis based on the similarity corresponding to each of the candidate subgraphs in combination with the satellite fault knowledge graph to obtain a fault diagnosis result includes: Sort each of the candidate subgraphs according to the similarity, query the satellite fault knowledge graph to determine the bottom events in the fault report of the fault phenomenon corresponding to the candidate subgraph, and calculate the probability value of the bottom events. Form a fault diagnosis tree based on the probability values of each of the bottom events, and determine a fault report corresponding to each bottom event.

9. A satellite fault diagnosis device based on graph embedding, characterized in that, Including: A query statement acquisition module, configured to acquire a fault phenomenon query statement, analyze the fault phenomenon query statement, and determine fault elements; A subgraph similarity calculation module, configured to construct a virtual subgraph based on each of the fault elements, determine candidate subgraphs based on each of the fault elements in combination with a pre-constructed satellite fault knowledge graph, determine a first target vector corresponding to the virtual subgraph and second target vectors corresponding to each of the candidate subgraphs, and calculate the similarity between the first target vector and each of the second target vectors to obtain the similarity between each of the candidate subgraphs and the virtual subgraph; wherein, the first target vector and the second target vector are determined based on a graph embedding algorithm; A fault diagnosis module, configured to perform fault diagnosis based on the similarities corresponding to each of the candidate subgraphs in combination with the satellite fault knowledge graph to obtain a fault diagnosis result.

10. An electronic device, characterized in that, The electronic device includes: At least one processor, and a memory communicatively connected to the at least one processor; Wherein, the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the graph embedding-based satellite fault diagnosis method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a processor to implement the graph embedding-based satellite fault diagnosis method according to any one of claims 1-8 when executed.

Images