A software-defined satellite error correction method based on large model knowledge graph

Abstract

The present invention relates to a method for guiding software-defined satellite error correction based on a large model knowledge graph. By combining the reasoning results of the large model with the contextual information in the knowledge graph, accurate and efficient fault identification and correction can be achieved. The knowledge graph contains the relationship between the satellite subsystems, historical fault cases and correction solutions. When a fault is detected, the system can quickly locate the source of the fault and generate optimized correction suggestions based on similar historical cases and reasoning rules. The present invention can also dynamically update the knowledge graph and optimize the reasoning rules to enhance the adaptability of the system in different environments. This method not only improves the accuracy of fault diagnosis and correction efficiency, reduces the need for manual intervention, but also significantly shortens the fault handling time, and improves the operational reliability and task completion rate of the satellite system. The invention has broad application prospects and helps to improve satellite operation and maintenance efficiency and reduce costs.

Description

Technical Field

[0001] The present invention relates to the field related to satellite systems, and specifically to a method for correcting satellite errors using software defined based on a large model knowledge graph. Background Art

[0002] As the complexity of satellite systems increases, traditional satellite operation, maintenance and fault correction methods have become difficult to cope with the dynamically changing space environment and diverse mission requirements. During the on-orbit operation of satellite systems, various faults may occur due to the influence of external space environmental factors (such as space debris and radiation) and internal system failures (such as sensor failure and communication interruption). At present, fault repair mostly relies on ground station control personnel to take corrective measures based on telemetry data analysis. This method has problems such as long delay, low efficiency and poor adaptability.

[0003] The introduction of software-defined satellites (SDS) enables satellite systems to correct faults on-orbit through software updates. However, how to effectively use big data and artificial intelligence technologies to enhance the adaptive and self-healing capabilities of SDS remains an urgent problem to be solved. By combining large models (such as Transformer models) with professional satellite knowledge graphs, efficient self-correction and optimization of satellite systems can be achieved through intelligent reasoning and real-time decision support, thereby improving operational efficiency and lifespan. Summary of the invention

[0004] This paper proposes a software-defined satellite error correction method based on a large model knowledge graph. By building a professional knowledge graph covering the entire process of satellite design, manufacturing, operation and maintenance, and combining a large model for reasoning and decision support, when a satellite fails, fast and accurate error correction can be achieved through automated software updates. This method can effectively shorten the response time of satellite fault correction and improve the adaptive capability and self-repair efficiency of the satellite system.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for guiding software-defined satellite error correction based on a large model knowledge graph, comprising the following steps:

[0006] Step 1: Knowledge graph construction and integration: By collecting and integrating the satellite system’s design standards, manufacturing specifications, fault case libraries, telemetry data, and operation and maintenance records, a knowledge graph covering the key components of the satellite system is constructed. The content of the knowledge graph includes satellite subsystems, specifically the key entities, relational data, and fault models of the communication system, power system, and attitude control system.

[0007] Step 2: Large model training and application: Based on the Transformer large model algorithm, the information in the knowledge graph in step 1 is used to learn and model the complex behavior of the satellite system. The large model continuously learns historical data of satellite operation, predicts satellite problems and their potential causes, and generates correction plans.

[0008] Step 3: Real-time telemetry data collection and fault detection: During the operation of the satellite, the telemetry data sent back by the satellite is collected in real time and input into the fault diagnosis system driven by the large model in step 2. The fault diagnosis system combines the knowledge graph to make judgments and automatically identify potential problems and fault sources;

[0009] Step 4: Adaptive error correction decision generation: Based on the ground control center or autonomous reasoning results, the decision recommendations generated by the large model will be sent to the satellite system through the ground station. After receiving the instructions, the satellite will automatically perform software logic updates, correct faults or adjust system parameters to achieve rapid repair and self-optimization.

[0010] Step 5: Continuous learning and optimization: Based on new failure modes and environmental changes discovered during satellite operation, the knowledge graph will continue to expand, and the big model will continuously update and optimize its prediction capabilities and correction solutions through new data inputs to improve the accuracy and response speed of future fault handling.

[0011] Preferably, the steps and key points of knowledge graph construction and integration in step 1 are as follows:

[0012] Step 11: Data collection: The construction of the knowledge graph first requires extensive collection of data on various aspects of the satellite system, including the following data sources:

[0013] Design standards and manufacturing specifications: including design documents, manufacturing processes, material specifications, and quality control standards for satellites and their subsystems. The above data is used in the knowledge graph to provide a detailed description of the system structure and design requirements; Fault case library: collects historical satellite fault cases, including fault descriptions, occurrence conditions, impact range, fault cause analysis and solutions. The fault case library is used to model and infer the failure modes and correction methods of satellite systems; Telemetry data: including real-time data transmitted back by satellites during in-orbit operation, including temperature, pressure, voltage, power, position and speed parameters. Telemetry data is used to analyze the operating status of satellite systems and detect potential anomalies and faults; Operation and maintenance records: including maintenance logs, adjustment records, and software update history of satellites during their in-orbit period. The records are used to understand the actual operation and maintenance behavior of the system and provide support for fault diagnosis and prediction;

[0014] Step 12: Data integration and preprocessing: Clean and preprocess the data collected in step 11 to remove redundant, erroneous and inconsistent data to obtain clean data. Ensuring the quality and consistency of data is the basis for building a knowledge graph; standardize the data from different data sources in the clean data, unify the format and unit, and obtain standardized data;

[0015] Step 13, knowledge graph modeling: According to the structure and function of the satellite system, define the entities and relationships in the knowledge graph, that is, semantically annotate the standardized data in step 12, where the entities include the satellite as a whole, each subsystem, component and environmental factors; then build a fault model for the satellite system, define the characteristics, impact range, trigger conditions and correction methods of different fault types according to actual needs, and the fault model is used to infer and predict potential problems of the satellite system; finally, integrate the relationship data between entities in the semantically annotated data, and use the knowledge graph to understand the overall architecture of the system and the fault propagation path;

[0016] Step 14: Semantic reasoning of knowledge graph: Based on usage requirements, define a set of semantic reasoning rules to analyze the data in the knowledge graph.

[0017] Preferably, the specific contents of the large model training and application in step 2 are as follows:

[0018] Step 21, model selection: The selection of the large model is based on the Transformer architecture. The core components of the Transformer model include: multi-head self-attention mechanism. Through the self-attention mechanism, the model captures the dynamic behavior between the satellite subsystems, which is described by the following formula:

[0019]

[0020] in, is the satellite's operating state vector, is the failure history, is the corresponding correction plan, From the input sequence, is the length of the satellite vector, used to scale the calculation results;

[0021] Step 22: Model input and embedding layer design: The input of the Transformer model is time series data, including telemetry data of satellite subsystems and information in the knowledge graph. The format of the input data is: ,in Representative time The satellite state vector at time , is the length of the time series. In the embedding layer design, the time series data is mapped to a vector space of fixed dimension to ensure that the model can process data efficiently. The specific data embedding is expressed as: ,in Yes Input The embedding layer combines the historical telemetry data and failure mode data of satellite operations to ensure that the model can understand the global status of the satellite system;

[0022] Step 23, model training objectives and loss function: The purpose of training the Transformer model is to predict future faults and fault correction solutions based on the historical operation data of the satellite, including fault prediction and correction solution generation. The content is as follows: Assuming a fault prediction task, the model needs to Predicting future failure states :

[0023]

[0024] in, is the predicted fault state, ranging from , is the model parameter, and the loss function uses cross entropy loss to measure the difference between the predicted value and the true label:

[0025]

[0026] Then, the model generates a correction plan, which is a multi-classification task to generate a suitable correction plan. , where the loss function is another cross entropy loss:

[0027]

[0028] in, is the number of correction options, It is The true label of the class scheme, is the probability predicted by the model, and the final loss of the model is the weighted sum of the failure prediction loss and the correction solution generation loss:

[0029]

[0030] in, and is a weight parameter used to balance the losses of the two tasks;

[0031] Step 24: Model prediction and reasoning: During satellite operation, the model performs reasoning based on real-time telemetry data. Through the multi-head self-attention mechanism, the model predicts faults based on the current state and historical information:

[0032]

[0033] At the same time, the model generates the corresponding correction plan:

[0034]

[0035] in, and They are models used for fault prediction and correction solution generation respectively.

[0036] Preferably, the specific steps of real-time telemetry data collection and fault detection in step 3 are as follows:

[0037] Step 31: Real-time telemetry data collection: The satellite sends telemetry data to the ground station during operation. The telemetry data contains the operating parameters of each satellite subsystem. The telemetry data is collected and transmitted back to the ground at a high frequency. There is a satellite state vector , which contains all monitoring parameters: ;

[0038] Step 32, data preprocessing and standardization: Before the telemetry data enters the fault detection system, data preprocessing is performed, including: missing value processing: for sensors that do not provide data due to faults, missing value filling or interpolation is performed; after completing the missing value processing, the telemetry data is denoised using a filtering method; after completing the data denoising, the telemetry data is standardized due to the different value ranges of the outputs of various sensors. The standardized telemetry data is in the form of: ;

[0039] Step 33: Data input into the big model: real-time telemetry data after preprocessing As input, it is fed into the Transformer-based large model. The model analyzes the time series data to determine whether there is a potential fault. The input sequence is: ,in, is the size of the time window, the model uses the past The data of each time step is used to infer the current state. The input of each time step is mapped to a high-dimensional space through the embedding layer: ;

[0040] Step 34: Self-attention mechanism for state analysis: The Transformer large model uses a multi-head self-attention mechanism to analyze the time series data in step 33 and calculate the correlation between each time step. That is, the model captures the dynamic behavior between the satellite subsystems: ;

[0041] Step 35: Fault Detection: Model Input Sequence Perform inference and determine whether the satellite has potential failures at each time step , the model outputs the current system fault state prediction ; The loss function of fault detection uses cross entropy loss to measure the difference between the model prediction value and the actual fault label: ;

[0042] Step 35: Knowledge graph combination and judgment: The prediction and reasoning results of the model in the above steps will be combined with the information in the knowledge graph to improve the accuracy of fault diagnosis, that is, to obtain accurate satellite diagnosis information;

[0043] Step 36, real-time fault location and report generation: After obtaining the satellite diagnostic information, the system automatically generates a fault report, indicating the potential fault source and the affected subsystems. The fault report will be generated by the following formula, combining the model output and knowledge graph information:

[0044]

[0045] in is the function that generates the report, which combines the probability of failure detected by the model and information in the knowledge graph.

[0046] Preferably, the specific steps of generating the adaptive error correction decision in step 4 are as follows:

[0047] Step 41, data input and fault detection: First, the system detects the potential fault of the satellite system in step 3 by continuously monitoring the real-time telemetry data, combining model reasoning and knowledge graph analysis. The detected fault triggers the adaptive error correction process. The output of the fault detection is: ;

[0048] Step 42, knowledge graph query and associated information extraction: After detecting a fault, the system queries the knowledge graph for information related to the fault. The system matches the fault type with the relevant entities and relationships in the knowledge graph and extracts correction suggestions, where the knowledge graph query is represented as: , It is the information related to the current fault extracted from the knowledge graph, including historical cases and correction solutions;

[0049] Step 43, model reasoning generates correction solutions: By combining the context information of the knowledge graph, the model infers and generates correction decisions based on the current fault status and the real-time data of the satellite. The correction decisions include specific operation instructions, specifically adjusting hardware components, modifying the satellite's on-orbit operation logic and software update instructions. The result of model reasoning is generated by multi-task learning, that is, predicting faults and correction solutions at the same time. The generation of the correction solution is expressed as: , It is a model function for generating correction decisions, which combines the current state of the satellite, knowledge graph information and model parameters; the multi-task loss function of the correction scheme is the cross entropy loss of the correction decision generation task: ;

[0050] Step 44, dynamic optimization and adaptive execution: After the system executes the operation according to the correction decision, it continuously collects telemetry data and observes the changes in the system status; if the correction is effective, the system will continue to run; if the fault is not completely repaired, the system will dynamically adjust the strategy and generate a new correction decision until the fault is completely resolved. That is, the model continuously optimizes the correction plan through the feedback mechanism so that it can adapt to different fault scenarios. Each correction execution will generate new data, and the feedback mechanism can be expressed as:

[0051]

[0052]

[0053] in, is the new state of the satellite after the correction is performed, It is the model's further prediction of failures and corrections in the new state.

[0054] Preferably, the specific steps of error correction and software update in step 5 are as follows:

[0055] Step 51, the ground control center and the generation of autonomous reasoning results: First, the ground control center analyzes the satellite operation status by receiving satellite telemetry data; at the same time, the model inside the satellite performs autonomous reasoning based on real-time data and knowledge graphs to generate correction decision suggestions, where the ground station uses the received telemetry data and the existing historical fault model to generate correction decisions; the satellite model uses real-time data input and knowledge graph information to generate correction plans and directly execute them. These two methods can be used separately or in combination. The decision suggestions of these two methods are expressed as follows through the reasoning results of the model: ;

[0056] Step 52, generation and issuance of instructions for correction decisions: When the ground control center or the satellite model generates correction suggestions, these decisions are converted into specific instruction sets and sent to the satellite system through the ground station;

[0057] Step 53, satellite receiving command and verification: the satellite system receives the command from the ground station through the communication link, and performs verification and verification to ensure data integrity and correctness;

[0058] Step 54, Feedback and Verification: After executing the command and correction process, the satellite system will feed back the updated and corrected results to the ground control center, which will verify the validity of the correction results to ensure that the fault is fixed and no new problems are generated;

[0059] Step 55, continuous learning and optimization: As the satellite continues to operate in orbit, the system accumulates more historical data and fault cases. The model gradually improves its fault detection and correction capabilities through continuous learning and dynamic updating of the knowledge graph. The process of continuous learning of the model is expressed as follows:

[0060]

[0061] is the learning rate, It is a comprehensive loss function that includes fault detection and correction decisions.

[0062] Preferably, an electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor:

[0063] The processor executes a method for guiding software-defined satellite error correction based on a large model knowledge graph by calling a computer program stored in the memory.

[0064] Preferably, a computer program product stored on a computer-readable medium includes a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement a software-defined satellite error correction method guided by a large model knowledge graph.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] Improve fault diagnosis accuracy: By combining the reasoning results of the big model with the contextual information in the knowledge graph, the present invention can quickly identify the fault type and source. The relationship between subsystems, historical fault cases and correction solutions contained in the knowledge graph provide rich background support for the big model, allowing the system to more accurately determine the cause of the fault and reduce the probability of misjudgment and missed judgment.

[0067] Accelerate fault response speed: By combining the real-time reasoning ability of large models with the knowledge accumulation of knowledge graphs, the present invention can quickly locate the source of faults and formulate correction plans at the early stage of fault occurrence. Using the fault propagation path information in the knowledge graph, the system can predict possible secondary faults, so as to take measures in advance and reduce system downtime and task interruption risks.

[0068] Enhance system adaptability: The present invention can dynamically update the knowledge graph and optimize the inference rules of the large model according to the new data and failure modes collected during the satellite operation, thereby enhancing the system's adaptability. The system can automatically adjust the fault diagnosis and correction strategy under different operating environments, such as temperature changes, radiation enhancement, etc., to ensure the continuous and stable operation of the satellite.

[0069] Optimize fault correction plan: By comparing the preliminary correction suggestions generated by the large model with historical cases in the knowledge graph, the present invention can further optimize the current correction plan. Combining the reasoning rules and environmental association data of the knowledge graph, the system can generate more accurate and effective correction instructions, improving the success rate and execution efficiency of fault correction.

[0070] Improve system operation efficiency and reliability: Through the dynamic expansion of knowledge graphs and continuous optimization of large models, the present invention enables satellite systems to maintain efficient operation in complex and ever-changing space environments. The system can quickly learn and adapt to new fault modes, improve the accuracy and speed of fault handling, and thus improve the completion rate of satellite missions and the reliability of operation.

[0071] Reduced operation and maintenance costs: Through automated fault detection, diagnosis and correction capabilities, the present invention reduces the need for manual intervention and reduces the cost of satellite operation and maintenance. The system can independently complete fault analysis and optimization operations, reducing the frequent intervention of ground stations on satellites and improving operation and maintenance efficiency.

[0072] In summary, the present invention combines the knowledge graph with the big model to form an efficient and intelligent satellite fault diagnosis and optimization system, which has significant practical value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0074] Figure 2 It is a flowchart of the large model training and application method of the present invention;

[0075] Figure 3 It is a flow chart of the real-time telemetry data collection and fault detection method of the present invention;

[0076] Figure 4 The figure is a schematic diagram of the structure of an electronic device. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] Example 1

[0079] See also Figure 1-3 The present invention provides a technical solution: a method for guiding software-defined satellite error correction based on a large model knowledge graph, comprising the following steps:

[0080] Step 1: Knowledge graph construction and integration: By collecting and integrating the satellite system’s design standards, manufacturing specifications, fault case libraries, telemetry data, and operation and maintenance records, a knowledge graph covering the key components of the satellite system is constructed. The content of the knowledge graph includes satellite subsystems, specifically the key entities, relational data, and fault models of the communication system, power system, and attitude control system.

[0081] Specifically, the steps and key points of knowledge graph construction and integration are as follows:

[0082] Step 11: Data collection: The construction of the knowledge graph first requires extensive collection of data on various aspects of the satellite system, including the following data sources:

[0083] Design standards and manufacturing specifications: including design documents, manufacturing processes, material specifications, and quality control standards for satellites and their subsystems. The above data is used in the knowledge graph to provide a detailed description of the system structure and design requirements; Fault case library: collects historical satellite fault cases, including fault descriptions, occurrence conditions, impact range, fault cause analysis and solutions. The fault case library is used to model and infer the failure modes and correction methods of satellite systems; Telemetry data: including real-time data transmitted back by satellites during in-orbit operation, including temperature, pressure, voltage, power, position and speed parameters. Telemetry data is used to analyze the operating status of satellite systems and detect potential anomalies and faults; Operation and maintenance records: including maintenance logs, adjustment records, and software update history of satellites during their in-orbit period. The records are used to understand the actual operation and maintenance behavior of the system and provide support for fault diagnosis and prediction;

[0084] Step 12: Data integration and preprocessing: Clean and preprocess the data collected in step 11 to remove redundant, erroneous and inconsistent data to obtain clean data. Ensuring the quality and consistency of data is the basis for building a knowledge graph; standardize the data from different data sources in the clean data, unify the format and unit, and obtain standardized data;

[0085] Step 13, knowledge graph modeling: According to the structure and function of the satellite system, define the entities and relationships in the knowledge graph, that is, semantically annotate the standardized data in step 12, where the entities include the satellite as a whole, each subsystem, component and environmental factors; then build a fault model for the satellite system, define the characteristics, impact range, trigger conditions and correction methods of different fault types according to actual needs, and the fault model is used to infer and predict potential problems of the satellite system; finally, integrate the relationship data between entities in the semantically annotated data, and use the knowledge graph to understand the overall architecture of the system and the fault propagation path;

[0086] It should be noted here that:

[0087] The entities of the knowledge graph represent different physical or logical objects in the satellite system. According to the structure and function of the satellite system, common entity definitions are as follows:

[0088] Satellite: The satellite itself is the top-level entity, representing the operating status and mission objectives of the entire system.

[0089] Subsystems: The main subsystems of a satellite, such as the communication system, power system, attitude control system, thermal control system, etc. Each subsystem performs a specific function.

[0090] Components: Specific hardware components of each subsystem, such as antennas in the communication system, battery packs in the power system, gyroscopes in the attitude control system, etc.

[0091] Environmental Factors: External conditions of satellite operation, such as space radiation, temperature changes, space debris, solar radiation, etc. These factors will affect the normal operation of the satellite;

[0092] The definition and classification of entities in the knowledge graph are:

[0093] Satellite Entity: The top-level entity that describes the overall status and mission of the satellite.

[0094] Subsystem: A mid-level entity that describes the function and operation of a subsystem.

[0095] Component: A low-level entity that describes the specific state and operating conditions of a hardware or software component.

[0096] Environmental factors: external entities that describe the effects of external influences on the satellite.

[0097] Definition of relationships between entities

[0098] Defining the relationship between entities is an important step in building a knowledge graph. Relationships reflect the interactions, physical connections, and influences between different entities. Common types of relationships include:

[0099] Physical connection relationship: For example, the physical connection relationship between components describes how subsystems and components are connected through signal cables, data buses, etc. Such as the signal transmission path between the antenna and the communication module, and the power supply link between the battery and the power system.

[0100] Control link: defines the control path between components. For example, how a satellite attitude control system adjusts the attitude of a satellite through a reaction flywheel or control software.

[0101] Dependencies: Some components or subsystems rely on other components to function properly. For example, the antenna of the communication system depends on the battery of the power system to function properly, and the attitude control system depends on the accurate sensing of the gyroscope.

[0102] Fault propagation path: defines other components or subsystems that may be affected when a fault occurs. For example, a battery failure may cause insufficient power supply, which in turn affects the response speed of the attitude control system.

[0103] Through these relationships, the knowledge graph can model the interdependencies and interactions of various subsystems and components in the satellite system.

[0104] The fault model is a key part of the knowledge graph, which is used to describe different types of faults and their characteristics, impact range, trigger conditions and correction methods. The fault model combined with the entities and relationships in the knowledge graph can help the system infer potential faults and find appropriate correction methods.

[0105] The main elements of the fault model include:

[0106] Fault Type: Different fault types, such as communication interruption, insufficient power supply, attitude control failure, etc. Each fault type is associated with one or more subsystems or components.

[0107] Fault characteristics: Characteristic signals that describe the fault, such as abnormal voltage, overtemperature, signal loss, etc. These characteristics can help the system identify the specific type of fault.

[0108] Trigger conditions: define the conditions under which a fault occurs. For example, a temperature exceeding a certain threshold may cause a hardware overheating fault, and a battery charge below a certain level may cause the power supply system to fail.

[0109] Impact scope: Defines the subsystems or components that a failure may affect. For example, a power supply system failure may affect multiple subsystems such as communications and attitude control.

[0110] Correction method: Provide corresponding correction solutions for each fault, including hardware adjustment, software update, system parameter optimization, etc. For example, a communication fault may be corrected by adjusting the antenna angle or increasing the transmission power.

[0111] The fault model not only describes the characteristics and impact of the fault, but also associates it with the entities and relationships in the knowledge graph to help the system quickly locate the problem and propose solutions when a fault occurs.

[0112] Relational data integration: When building a knowledge graph, the integration of relational data is crucial. By integrating various relational data in the satellite system, the knowledge graph can better understand the overall architecture of the system and the fault propagation path. The steps to integrate relational data include:

[0113] Physical connection data integration: Integrate the physical connection relationships between subsystems and components, such as the signal transmission path between the antenna and the communication module, the connection between the battery and the power management system, etc. These connection relationships help understand the status of the system at runtime and the possible impact area when a fault occurs.

[0114] Control link integration: Integrate the control paths and data flows of different subsystems in the system, such as feedback loops in attitude control systems, signal transmission links in communication systems, etc. These relationships can help the system understand the transmission and processing of data between different components, making it easier to locate faults.

[0115] Fault propagation path integration: Integrate known fault propagation paths into the knowledge graph. For example, when a battery has a problem, which subsystems will be affected? This helps to quickly identify related subsystems and components when a fault occurs and make corrections.

[0116] Step 14: Semantic reasoning of knowledge graph: Based on usage requirements, define a set of semantic reasoning rules to analyze the data in the knowledge graph.

[0117] The knowledge graph constructed through the steps of knowledge graph construction and integration can provide comprehensive knowledge support for the operation and maintenance of satellite systems. Combined with the reasoning and decision-making capabilities of large models, the knowledge graph can significantly improve the adaptive error correction capabilities of software-defined satellites and provide strong guarantees for the efficient and stable operation of satellite systems.

[0118] Step 2: Large model training and application: Based on the Transformer large model algorithm, the information in the knowledge graph in step 1 is used to learn and model the complex behavior of the satellite system. The large model continuously learns historical data of satellite operation, predicts satellite problems and their potential causes, and generates correction plans.

[0119] Specifically, the specific contents of large model training and application are as follows:

[0120] Step 21, model selection: The selection of the large model is based on the Transformer architecture. The core components of the Transformer model include: multi-head self-attention mechanism. Through the self-attention mechanism, the model captures the dynamic behavior between the satellite subsystems, which is described by the following formula:

[0121]

[0122] in, is the satellite's operating state vector, is the failure history, is the corresponding correction plan, From the input sequence, is the length of the satellite vector, used to scale the calculation results;

[0123] Step 22: Model input and embedding layer design: The input of the Transformer model is time series data, including telemetry data of satellite subsystems and information in the knowledge graph. The format of the input data is: ,in Representative time The satellite state vector at time , is the length of the time series. In the embedding layer design, the time series data is mapped to a vector space of fixed dimension to ensure that the model can process data efficiently. The specific data embedding is expressed as: ,in Yes Input The embedding layer combines the historical telemetry data and failure mode data of satellite operations to ensure that the model can understand the global status of the satellite system;

[0124] Step 23, model training objectives and loss function: The purpose of training the Transformer model is to predict future faults and fault correction solutions based on the historical operation data of the satellite, including fault prediction and correction solution generation. The content is as follows: Assuming a fault prediction task, the model needs to Predicting future failure states :

[0125]

[0126] in, is the predicted fault state, ranging from , is the model parameter, and the loss function uses cross entropy loss to measure the difference between the predicted value and the true label:

[0127]

[0128] Then, the model generates a correction plan, which is a multi-classification task to generate a suitable correction plan. , where the loss function is another cross entropy loss:

[0129]

[0130] in, is the number of correction options, It is The true label of the class scheme, is the probability predicted by the model, and the final loss of the model is the weighted sum of the failure prediction loss and the correction solution generation loss:

[0131]

[0132] in, and is a weight parameter used to balance the losses of the two tasks;

[0133] Step 24: Model prediction and reasoning: During satellite operation, the model performs reasoning based on real-time telemetry data. Through the multi-head self-attention mechanism, the model predicts faults based on the current state and historical information:

[0134]

[0135] At the same time, the model generates the corresponding correction plan:

[0136]

[0137] in, and They are models used for fault prediction and correction solution generation respectively.

[0138] Through the Transformer-based large model, combined with the knowledge graph and historical telemetry data of the satellite system, the model can achieve fault prediction and automatic correction plan generation. Through the self-attention mechanism, the model can make full use of historical information and continuously improve its accuracy and efficiency in satellite system fault correction through continuous learning and optimization.

[0139] Step 3: Real-time telemetry data collection and fault detection: During the operation of the satellite, the telemetry data sent back by the satellite is collected in real time and input into the fault diagnosis system driven by the large model in step 2. The fault diagnosis system combines the knowledge graph to make judgments and automatically identify potential problems and fault sources;

[0140] Specifically, the specific steps for real-time telemetry data collection and fault detection are as follows:

[0141] Step 31, real-time telemetry data collection: The satellite sends telemetry data to the ground station during operation. The telemetry data contains the operating parameters of each satellite subsystem, such as:

[0142] Physical parameters such as temperature, pressure, power, voltage, current, etc.

[0143] Attitude control information (such as angular velocity, angle, etc.);

[0144] Communication status (such as signal strength, transmission delay, etc.);

[0145] The above telemetry data is collected and transmitted back to the ground at a high frequency. There is a satellite state vector , which contains all monitoring parameters: ;

[0146] Step 32, data preprocessing and standardization: Before the telemetry data enters the fault detection system, data preprocessing is performed, including: missing value processing: for sensors that do not provide data due to faults, missing value filling or interpolation is performed; after completing the missing value processing, the telemetry data is denoised using a filtering method; after completing the data denoising, the telemetry data is standardized due to the different value ranges of the outputs of various sensors. The standardized telemetry data is in the form of: ;

[0147] Step 33: Data input into the big model: real-time telemetry data after preprocessing As input, it is fed into the Transformer-based large model. The model analyzes the time series data to determine whether there is a potential fault. The input sequence is: ,in, is the size of the time window, the model uses the past The data of each time step is used to infer the current state. The input of each time step is mapped to a high-dimensional space through the embedding layer: ;

[0148] Step 34: Self-attention mechanism for state analysis: The Transformer large model uses a multi-head self-attention mechanism to analyze the time series data in step 33 and calculate the correlation between each time step. That is, the model captures the dynamic behavior between the satellite subsystems: ;

[0149] Step 35: Fault Detection: Model Input Sequence Perform inference and determine whether the satellite has potential failures at each time step , the model outputs the current system fault state prediction ; The loss function of fault detection uses cross entropy loss to measure the difference between the model prediction value and the actual fault label: ;

[0150] Step 35, knowledge graph combination and judgment: The prediction and reasoning results of the model in the above steps will be combined with the information in the knowledge graph to improve the accuracy of fault diagnosis, that is, to obtain accurate satellite diagnosis information; the knowledge graph contains the relationship between the satellite subsystems, historical fault cases and correction solutions. For example:

[0151] If the model detects a drop in battery voltage and similar fault conditions are recorded in the knowledge graph (such as voltage instability caused by an overload in the communications system), the system can quickly locate the source of the fault.

[0152] Combined with the reasoning rules of the knowledge graph, the diagnosis can be further refined. Assuming that the communication signal strength in the telemetry data suddenly decreases, the knowledge graph may indicate that it is related to the antenna position or power supply status, thus providing guidance for the model's fault detection.

[0153] Step 36, real-time fault location and report generation: After obtaining the satellite diagnostic information, the system automatically generates a fault report, indicating the potential fault source and the affected subsystems. The fault report will be generated by the following formula, combining the model output and knowledge graph information:

[0154]

[0155] in is the function that generates the report, which combines the probability of failure detected by the model and information in the knowledge graph.

[0156] By collecting satellite telemetry data in real time, combining the deep learning capabilities of large models with the contextual information of knowledge graphs, the system can efficiently detect potential faults in satellite operation and automatically generate fault reports and correction suggestions. This process uses the self-attention mechanism of Transformer to analyze time series data, and through the reasoning support of knowledge graphs, it realizes intelligent fault diagnosis of complex systems.

[0157] Step 4: Adaptive error correction decision generation: Based on the ground control center or autonomous reasoning results, the decision recommendations generated by the large model will be sent to the satellite system through the ground station. After receiving the instructions, the satellite will automatically perform software logic updates, correct faults or adjust system parameters to achieve rapid repair and self-optimization.

[0158] Specifically, the specific steps for adaptive error correction decision generation are as follows:

[0159] Step 41, data input and fault detection: First, the system detects the potential fault of the satellite system in step 3 by continuously monitoring the real-time telemetry data, combining model reasoning and knowledge graph analysis. The detected fault triggers the adaptive error correction process. The output of the fault detection is: ;

[0160] Step 42, knowledge graph query and associated information extraction: After a fault is detected, the system queries the knowledge graph for information related to the fault, matches the fault type with the related entities and relationships in the knowledge graph, and extracts correction suggestions;

[0161] Assuming that the detected fault is signal attenuation of the satellite communication system, the system will query the knowledge graph for entities and relationships related to "communication system", for example:

[0162] Subcomponents related to communication systems (such as antennas, power modules);

[0163] Correction plans for similar faults in history (such as adjusting the antenna direction, increasing power, etc.);

[0164] The knowledge graph query is expressed as: , It is the information related to the current fault extracted from the knowledge graph, including historical cases and correction solutions;

[0165] Step 43, model reasoning generates correction solutions: By combining the context information of the knowledge graph, the model infers and generates correction decisions based on the current fault status and the real-time data of the satellite. The correction decisions include specific operation instructions, specifically adjusting hardware components, modifying the satellite's on-orbit operation logic and software update instructions. The result of model reasoning is generated by multi-task learning, that is, predicting faults and correction solutions at the same time. The generation of the correction solution is expressed as: , It is a model function for generating correction decisions, which combines the current state of the satellite, knowledge graph information and model parameters; the multi-task loss function of the correction scheme is the cross entropy loss of the correction decision generation task: ;

[0166] Step 44, dynamic optimization and adaptive execution: After the system executes the operation according to the correction decision, it continuously collects telemetry data and observes the changes in the system status; if the correction is effective, the system will continue to run; if the fault is not completely repaired, the system will dynamically adjust the strategy and generate a new correction decision until the fault is completely resolved. That is, the model continuously optimizes the correction plan through the feedback mechanism so that it can adapt to different fault scenarios. Each correction execution will generate new data, and the feedback mechanism can be expressed as:

[0167]

[0168]

[0169] in, is the new state of the satellite after the correction is performed, It is the model's further prediction of failures and corrections in the new state.

[0170] The adaptive error correction decision generation process can effectively identify potential satellite failures and generate targeted correction plans through the close integration of large models and knowledge graphs. The system can not only dynamically adjust hardware and software, but also improve its own fault repair capabilities through continuous learning to ensure the long-term stable operation of the satellite system in orbit.

[0171] Step 5: Continuous learning and optimization: Based on new failure modes and environmental changes discovered during satellite operation, the knowledge graph will continue to expand, and the big model will continuously update and optimize its prediction capabilities and correction solutions through new data inputs to improve the accuracy and response speed of future fault handling.

[0172] Specifically, the steps for error correction and software update are as follows:

[0173] Step 51, the ground control center and the generation of autonomous reasoning results: First, the ground control center analyzes the satellite operation status by receiving satellite telemetry data; at the same time, the model inside the satellite performs autonomous reasoning based on real-time data and knowledge graphs to generate correction decision suggestions, where the ground station uses the received telemetry data and the existing historical fault model to generate correction decisions; the satellite model uses real-time data input and knowledge graph information to generate correction plans and directly execute them. These two methods can be used separately or in combination. The decision suggestions of these two methods are expressed as follows through the reasoning results of the model: ;

[0174] Step 52, generation and issuance of instructions for correction decisions: When the ground control center or the satellite model generates correction suggestions, these decisions are converted into specific instruction sets and sent to the satellite system through the ground station;

[0175] Step 53, satellite receiving command and verification: the satellite system receives the command from the ground station through the communication link, and performs verification and verification to ensure data integrity and correctness;

[0176] Step 54, Feedback and Verification: After executing the command and correction process, the satellite system will feed back the updated and corrected results to the ground control center, which will verify the validity of the correction results to ensure that the fault is fixed and no new problems are generated;

[0177] Step 55, continuous learning and optimization: As the satellite continues to operate in orbit, the system accumulates more historical data and fault cases. The model gradually improves its fault detection and correction capabilities through continuous learning and dynamic updating of the knowledge graph. The process of continuous learning of the model is expressed as follows:

[0178]

[0179] is the learning rate, It is a comprehensive loss function that includes fault detection and correction decisions.

[0180] Based on the ground control center and satellite autonomous reasoning, the correction suggestions generated by the large model can quickly repair satellite faults and optimize parameters through command issuance and software updates. Through continuous monitoring and feedback, the system can continuously optimize itself and improve its stability and efficiency in orbit.

[0181] Example 2

[0182] An electronic device according to an exemplary embodiment includes: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0183] The processor executes the above-mentioned method of guiding software-defined satellite error correction based on a large model knowledge graph by calling the computer program stored in the memory.

[0184] Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. The electronic device may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) and one or more memories, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement a software-defined satellite error correction method based on a large model knowledge graph guided by the above-mentioned various method embodiments.

[0185] The electronic device may also include other components for realizing the functions of the device, for example, the electronic device may also have components such as a wired or wireless network interface and an input / output interface for input and output. The embodiments of the present application will not be described in detail here.

[0186] This embodiment also provides a computer program product stored on a computer-readable medium, including a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the method for guiding software-defined satellite error correction based on a large model knowledge graph.

[0187] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0188] It should be understood that determining B based on A does not mean determining B only based on A. B can also be determined based on A and / or other information.

[0189] The present invention achieves the following effects: improving fault diagnosis accuracy: utilizing historical fault cases and system relationships in the knowledge graph, combined with large model reasoning, can accurately locate the source of the fault and reduce misjudgment; accelerating fault processing: quickly analyzing the fault propagation path, predicting secondary faults in advance, and shortening the fault correction time; optimizing correction plans: optimizing correction suggestions based on the knowledge graph to improve the effectiveness of the correction plan; reducing operation and maintenance costs: reducing manual intervention, realizing automated fault diagnosis and repair, and improving operation and maintenance efficiency.

[0190] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A software-defined satellite error correction method based on a large model knowledge graph, characterized in that: The following steps are involved: Step 1: Knowledge graph construction and integration: By collecting and integrating the satellite system’s design standards, manufacturing specifications, fault case libraries, telemetry data, and operation and maintenance records, a knowledge graph covering the key components of the satellite system is constructed. The content of the knowledge graph includes satellite subsystems, specifically the key entities, relational data, and fault models of the communication system, power system, and attitude control system. Step 2: Large model training and application: Based on the Transformer large model algorithm, the information in the knowledge graph in step 1 is used to learn and model the complex behavior of the satellite system. The large model continuously learns historical data of satellite operation, predicts satellite problems and their potential causes, and generates correction plans. The specific contents of the large model training and application in step 2 are as follows: Step 21, model selection: The selection of the large model is based on the Transformer architecture. The core components of the Transformer model include: multi-head self-attention mechanism. Through the self-attention mechanism, the model captures the dynamic behavior between the satellite subsystems, which is described by the following formula: in, is the satellite's operating state vector, is the failure history, is the corresponding correction plan, From the input sequence, is the length of the satellite vector, used to scale the calculation results; Step 22: Model input and embedding layer design: The input of the Transformer model is time series data, including telemetry data of satellite subsystems and information in the knowledge graph. The format of the input data is: ,in Representative time The satellite state vector at time , is the length of the time series. In the embedding layer design, the time series data is mapped to a vector space of fixed dimension to ensure that the model can process data efficiently. The specific data embedding is expressed as: ,in Yes Input The embedding layer combines the historical telemetry data and failure mode data of satellite operations to ensure that the model can understand the global status of the satellite system; Step 23, model training objectives and loss function: The purpose of training the Transformer model is to predict future faults and fault correction solutions based on the historical operation data of the satellite, including fault prediction and correction solution generation. The content is as follows: Assuming a fault prediction task, the model needs to Predicting future failure states : in, is the predicted fault state, ranging from , is the model parameter, and the loss function uses cross entropy loss to measure the difference between the predicted value and the true label: Then, the model generates a correction plan, which is a multi-classification task to generate a suitable correction plan. , where the loss function is another cross entropy loss: in, is the number of correction options, It is The true label of the class scheme, is the probability predicted by the model, and the final loss of the model is the weighted sum of the failure prediction loss and the correction solution generation loss: in, and is a weight parameter used to balance the losses of the two tasks; Step 24: Model prediction and reasoning: During satellite operation, the model performs reasoning based on real-time telemetry data. Through the multi-head self-attention mechanism, the model predicts faults based on the current state and historical information: At the same time, the model generates the corresponding correction plan: in, and They are the models used for fault prediction and correction solution generation respectively; Step 3: Real-time telemetry data collection and fault detection: During the operation of the satellite, the telemetry data sent back by the satellite is collected in real time and input into the fault diagnosis system driven by the large model in step 2. The fault diagnosis system combines the knowledge graph to make judgments and automatically identify potential problems and fault sources; The specific steps of real-time telemetry data collection and fault detection in step 3 are as follows: Step 31: Real-time telemetry data collection: The satellite sends telemetry data to the ground station during operation. The telemetry data contains the operating parameters of each satellite subsystem. The telemetry data is collected and transmitted back to the ground at a high frequency. There is a satellite state vector , which contains all monitoring parameters: ; Step 32, data preprocessing and standardization: Before the telemetry data enters the fault detection system, data preprocessing is performed, including: missing value processing: for sensors that do not provide data due to faults, missing value filling or interpolation is performed; after completing the missing value processing, the telemetry data is denoised using a filtering method; after completing the data denoising, the telemetry data is standardized due to the different value ranges of the outputs of various sensors. The standardized telemetry data is in the form of: ; Step 33: Data input into the big model: real-time telemetry data after preprocessing As input, it is fed into the Transformer-based large model. The model analyzes the time series data to determine whether there is a potential fault. The input sequence is: ,in, is the size of the time window, the model uses the past The data of each time step is used to infer the current state. The input of each time step is mapped to a high-dimensional space through the embedding layer: ; Step 34: Self-attention mechanism for state analysis: The Transformer large model uses a multi-head self-attention mechanism to analyze the time series data in step 33 and calculate the correlation between each time step. That is, the model captures the dynamic behavior between the satellite subsystems: ; Step 35: Fault Detection: Model Input Sequence Perform inference and determine whether the satellite has potential failures at each time step , the model outputs the current system fault state prediction ; The loss function of fault detection uses cross entropy loss to measure the difference between the model prediction value and the actual fault label: ; Step 36: Knowledge graph combination and judgment: The prediction and reasoning results of the model in the above steps will be combined with the information in the knowledge graph to improve the accuracy of fault diagnosis, that is, to obtain accurate satellite diagnosis information; Step 37, real-time fault location and report generation: After obtaining satellite diagnostic information, the system automatically generates a fault report, indicating the potential fault source and affected subsystems. The fault report will be generated by the following formula, combining the model output and knowledge graph information: in is the function that generates the report, which combines the probability of failure detected by the model and information in the knowledge graph; Step 4: Adaptive error correction decision generation: Based on the ground control center or autonomous reasoning results, the decision recommendations generated by the large model will be sent to the satellite system through the ground station. After receiving the instructions, the satellite will automatically perform software logic updates, correct faults or adjust system parameters to achieve rapid repair and self-optimization. The specific steps of generating the adaptive error correction decision in step 4 are as follows: Step 41, data input and fault detection: First, the system detects the potential fault of the satellite system in step 3 by continuously monitoring the real-time telemetry data, combining model reasoning and knowledge graph analysis. The detected fault triggers the adaptive error correction process. The output of the fault detection is: ; Step 42, knowledge graph query and associated information extraction: After detecting a fault, the system queries the knowledge graph for information related to the fault. The system matches the fault type with the relevant entities and relationships in the knowledge graph and extracts correction suggestions, where the knowledge graph query is represented as: , It is the information related to the current fault extracted from the knowledge graph, including historical cases and correction solutions; Step 43, model reasoning generates correction solutions: By combining the context information of the knowledge graph, the model infers and generates correction decisions based on the current fault status and the real-time data of the satellite. The correction decisions include specific operation instructions, specifically adjusting hardware components, modifying the satellite's on-orbit operation logic and software update instructions. The result of model reasoning is generated by multi-task learning, that is, predicting faults and correction solutions at the same time. The generation of the correction solution is expressed as: , It is a model function for generating correction decisions, which combines the current state of the satellite, knowledge graph information and model parameters; the multi-task loss function of the correction scheme is the cross entropy loss of the correction decision generation task: ; Step 44, dynamic optimization and adaptive execution: After the system executes the operation according to the correction decision, it continuously collects telemetry data and observes the changes in the system status; if the correction is effective, the system will continue to run; if the fault is not completely repaired, the system will dynamically adjust the strategy and generate a new correction decision until the fault is completely resolved. That is, the model continuously optimizes the correction plan through the feedback mechanism so that it can adapt to different fault scenarios. Each correction execution will generate new data, and the feedback mechanism can be expressed as: in, is the new state of the satellite after the correction is performed, It is the model’s further prediction of failures and corrections in the new state; Step 5: Continuous learning and optimization: Based on new failure modes and environmental changes discovered during satellite operation, the knowledge graph will continue to expand, and the big model will continuously update and optimize its prediction capabilities and correction solutions through new data inputs, improving the accuracy and response speed of future fault handling; The specific steps of error correction and software update in step 5 are as follows: Step 51, the ground control center and the generation of autonomous reasoning results: First, the ground control center analyzes the satellite operation status by receiving satellite telemetry data; at the same time, the model inside the satellite performs autonomous reasoning based on real-time data and knowledge graphs to generate correction decision suggestions, where the ground station uses the received telemetry data and the existing historical fault model to generate correction decisions; the satellite model uses real-time data input and knowledge graph information to generate correction plans and directly execute them. These two methods can be used separately or in combination. The decision suggestions of these two methods are expressed as follows through the reasoning results of the model: ; Step 52, generation and issuance of instructions for correction decisions: When the ground control center or the satellite model generates correction suggestions, these decisions are converted into specific instruction sets and sent to the satellite system through the ground station; Step 53, satellite receiving command and verification: the satellite system receives the command from the ground station through the communication link, and performs verification and verification to ensure data integrity and correctness; Step 54, Feedback and Verification: After executing the command and correction process, the satellite system will feed back the updated and corrected results to the ground control center, which will verify the validity of the correction results to ensure that the fault is fixed and no new problems are generated; Step 55, continuous learning and optimization: As the satellite continues to operate in orbit, the system accumulates more historical data and fault cases. The model gradually improves its fault detection and correction capabilities through continuous learning and dynamic updating of the knowledge graph. The process of continuous learning of the model is expressed as follows: is the learning rate, It is a comprehensive loss function that includes fault detection and correction decisions.

2. According to claim 1, a method for guiding software-defined satellite error correction based on a large model knowledge graph is characterized by: The steps and key points of knowledge graph construction and integration in step 1 are as follows: Step 11: Data collection: The construction of the knowledge graph first requires extensive collection of data on various aspects of the satellite system, including the following data sources: Design standards and manufacturing specifications: including design documents, manufacturing processes, material specifications, and quality control standards for satellites and their subsystems. The above data is used in the knowledge graph to provide a detailed description of the system structure and design requirements; Fault case library: collects historical satellite fault cases, including fault descriptions, occurrence conditions, impact range, fault cause analysis and solutions. The fault case library is used to model and infer the failure modes and correction methods of satellite systems; Telemetry data: including real-time data transmitted back by satellites during in-orbit operation, including temperature, pressure, voltage, power, position and speed parameters. Telemetry data is used to analyze the operating status of satellite systems and detect potential anomalies and faults; Operation and maintenance records: including maintenance logs, adjustment records, and software update history of satellites during their in-orbit period. The records are used to understand the actual operation and maintenance behavior of the system and provide support for fault diagnosis and prediction; Step 12: Data integration and preprocessing: Clean and preprocess the data collected in step 11 to remove redundant, erroneous and inconsistent data to obtain clean data. Ensuring the quality and consistency of data is the basis for building a knowledge graph; standardize the data from different data sources in the clean data, unify the format and unit, and obtain standardized data; Step 13, knowledge graph modeling: According to the structure and function of the satellite system, define the entities and relationships in the knowledge graph, that is, semantically annotate the standardized data in step 12, where the entities include the satellite as a whole, each subsystem, component and environmental factors; then build a fault model for the satellite system, define the characteristics, impact range, trigger conditions and correction methods of different fault types according to actual needs, and the fault model is used to infer and predict potential problems of the satellite system; finally, integrate the relationship data between entities in the semantically annotated data, and use the knowledge graph to understand the overall architecture of the system and the fault propagation path; Step 14: Semantic reasoning of knowledge graph: Based on usage requirements, define a set of semantic reasoning rules to analyze the data in the knowledge graph.

3. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor: The processor executes the software-defined satellite error correction method guided by a large model knowledge graph as described in any one of claims 1 to 2 by calling the computer program stored in the memory.

4. A computer program product stored on a computer readable medium, characterized in that: It includes a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement a software-defined satellite error correction method based on a large model knowledge graph as described in any one of claims 1-2.

Images