A system and method for detecting anomalies in the orbital state of aircraft based on data association mining

Abstract

This invention relates to a system and method for detecting anomalies in the on-orbit status of aircraft based on data association mining. The system receives experimental data during aircraft missions and, according to the specific needs and characteristics of the evaluation mission scenario, performs data cleaning processes such as outlier removal, frequency and phase synchronization, and multi-source data fusion. The cleaned data is then normalized throughout the entire mission lifecycle. It can uniformly retrieve, store, manage, and share multi-source heterogeneous data throughout the entire lifecycle, featuring fast access speed and small storage capacity, and provides data visualization services based on the search catalog. Focusing on the characteristics of different experimental missions such as orbital safety inspection, orbital capability repair, on-orbit operation, and refueling, a knowledge base and rule base for aircraft experimental mission data are constructed. Data association mining and pattern analysis are then performed to provide technical support for fault diagnosis and anomaly detection of aircraft elements.

Description

Technical Field

[0001] This invention relates to the field of spacecraft on-orbit telemetry, tracking and command (TT&C) and data mining technology, and particularly to a system and method for detecting anomalies in the on-orbit status of spacecraft based on data association mining. Background Technology

[0002] With the development and application of modern high-tech technologies, especially aerospace technology, human activities have expanded from the Earth's surface to outer space. Space has become an important arena for human scientific exploration, space activities, and even future economic, social, and military activities. In space, the most important elements are the various spacecraft operating in orbit, providing services closely related to human economic and social life, such as navigation, communication, surveying, and environmental monitoring.

[0003] The development and launch of aircraft have accumulated a wealth of experimental data, including real-time / historical telemetry, telemetry, data transmission, data transmission images, and simulated flight results generated during missions such as long-distance approach, orbital rendezvous, close approach, hovering, evacuation, on-orbit maintenance, attitude and orbit takeover, and space operations. On the one hand, the sheer volume and complexity of this data pose significant challenges to data storage and subsequent use if not managed effectively and promptly. Therefore, ground-based systems need to organize and analyze this data to evaluate the experimental results. On the other hand, this data reflects the aircraft's inherent characteristics and flight status, which is crucial for analyzing aircraft performance and guiding the design of aircraft control systems. Researching data sharing and analysis techniques to obtain correlation rules and clustering envelopes between data, transforming chaotic and unstructured data into structured data with regular patterns, and extracting useful information can help improve aircraft performance. Simultaneously, as a valuable data resource, flight process data also needs to be managed and shared to support relevant units in conducting analysis, mining, and research, thereby improving the understanding of new problems and phenomena.

[0004] Meanwhile, big data has permeated many aspects of people's lives, and data mining has become a key focus in the field of big data processing. Data mining is the process of exploring valuable knowledge from large and diverse target datasets, mainly including association analysis, cluster analysis, classification analysis, and time series analysis. Among them, association rule mining is one of the most crucial branches of data mining, identifying and mining the relationships between data attributes in large datasets to help people make more effective and satisfactory decisions. Conducting research on data mining algorithms, establishing structured association rules between data, and using association mining algorithms to establish association descriptions between flight status and control parameters, obtaining their association strength, can optimize the spacecraft mission process, realize anomaly detection in the spacecraft's on-orbit status, and improve the overall performance of the spacecraft. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of complex and massive data, which can cause great difficulties in data storage and subsequent use, and to provide a system and method for detecting anomalies in the on-orbit status of aircraft based on data association mining.

[0006] In view of this, the present invention provides a system and method for detecting anomalies in the on-orbit status of spacecraft based on data association mining. The spacecraft on-orbit status anomaly detection system based on data association mining includes: Data preprocessing module: used to perform corresponding data preprocessing and data cleaning according to the characteristics of different experimental tasks; Data frequency and phase processing module: used to perform data frequency and phase adjustment based on the characteristics of different types, sources, frequencies and phases of data generated by different channels in the aircraft; Multi-source data fusion module: used to integrate, analyze, correlate, and fuse multiple types of information for observations of the same variable; Full-cycle data normalization module: Used to perform full-cycle data normalization processing for three types of data: structured, semi-structured, and unstructured, so as to convert data values ​​to the same units and orders of magnitude; The data association mining module is used to analyze the discrete state quantities and continuous analog quantities in telemetry parameters based on the characteristics of different experimental tasks. Discrete state quantities are represented by a combination of parameter names and state values, while continuous analog quantities are represented by trend symbols. This process unifies the construction of transaction sets and, through the construction of frequent pattern trees, mines the rules and knowledge between various associated parameters and operating states in the spacecraft telemetry parameters. The knowledge reasoning and anomaly detection module is used to establish anomaly judgment templates and rule bases based on the rules and knowledge obtained from the data association mining module for telemetry parameters and spacecraft test action sequences. It also formulates telemetry parameter rules and spacecraft test mission status rules. The telemetry parameter rules are used to specify the value range of telemetry parameters in different test mission states of the spacecraft, and the spacecraft test mission status rules are used to compare the action sequences of the spacecraft's on-orbit test missions to achieve spacecraft fault diagnosis.

[0007] As an improvement to the above system, the spacecraft on-orbit state anomaly detection system based on data association mining also includes: Full-cycle data management and sharing module: used for high-bandwidth storage, read / write, and role-based access control of multi-source heterogeneous data throughout the entire lifecycle of aircraft test missions; and Data Analysis Interactive Display Module: Used to perform analysis and display on full-cycle data, data mining results, knowledge reasoning and pattern analysis results, as well as interactive operations including data management and sharing, and manual data processing.

[0008] As an improvement to the above system, the data association mining module uses the FP-Growth association rule mining algorithm, and the specific execution steps are as follows: Step 1: Construct a transaction set using normalized aircraft telemetry parameters; Step 2: Scan the transaction set, generate frequent item sets and sort them in descending order of support, then store them in table L; Step 3: Create the root node of the FP tree, and recursively insert the items in each transaction according to the order of the L table to build a frequent pattern tree; Step 4: Mine frequent itemsets. Starting from the end of each branch of the FP-tree and working backwards, count the items that meet the minimum support and their frequency of occurrence. Step 5: Set minimum support and confidence thresholds, filter rules that meet the confidence level, and balance the redundancy rate of rule mining with the retention rate of known rules; Step 6: Evaluate the mining accuracy and redundancy rate, and adjust the confidence interval to retain potential unknown rules.

[0009] As an improvement to the above system, the knowledge reasoning and anomaly detection module performs the following steps: Step 1: Match the aircraft telemetry parameters with the anomaly interpretation template; Step 2: Perform dynamic weighting of spacecraft telemetry parameter templates; Step 3: Conduct anomaly detection and assessment of aircraft telemetry parameters; Step 4: If the evaluation is abnormal, proceed to step 5; otherwise, proceed to step 3 for the next spacecraft telemetry parameter until the traversal is complete. Step 5: Record abnormal states of the aircraft's telemetry parameters.

[0010] As an improvement to the above system, the system architecture is divided into 5 layers, including: Basic support layer: used for data sharing and analysis technology verification software, including computing server clusters, storage server clusters, and network devices; Data management layer: Used to manage various data resources that are stored, transferred, and generated, and to store multi-source heterogeneous data; Model resource layer: used for unified management of various data processing and computing models; Service scheduling layer: used to provide services for basic components, business middle platform, and data middle platform; and Application layer: Used to display applications, provide human-computer interaction interfaces, and form different application clients.

[0011] As an improvement to the aforementioned system, the data management layer manages various data resources including: real-time and historical telemetry, telemetry, data transmission, data transmission images, and simulated flight results from external ground simulation systems, as well as raw experimental data; data cleaning results after preprocessing such as data splicing, data interpolation or sampling, outlier removal, and format conversion; normalized results of the entire lifecycle of aircraft elements; prior rules and knowledge base data; results of pattern mining and knowledge reasoning analysis of multi-source experimental data; and results of fault diagnosis and abnormal state detection of aircraft elements. The data is stored in combination with a large-scale spatial database, database relational tables, and document cataloging library based on commercial databases. Relational, non-relational, and time-series storage modes are adopted for different data types and application characteristics to store multi-source heterogeneous data.

[0012] As an improvement to the above system, the data processing and computing model under unified management of the model resource layer includes: Data preprocessing models include data extraction, Lagrange interpolation model, and Dixon criterion model; Data synchronous frequency and phase processing model: including multi-channel data synchronous frequency and multi-channel data time synchronization; Multi-source data fusion models: including weighted data fusion models; Full-cycle data normalization model: including the time pyramid model; Data association mining models include state feature transformation analysis and FP-Growth association rule mining models; and Knowledge reasoning and anomaly detection models: including anomaly interpretation template matching and dynamic weighted models.

[0013] As an improvement to the above system, the service scheduling layer specifically includes: Basic components: These include common basic components such as data access components, message middleware components, microservice development framework, API gateway, task scheduling framework, and log center, which are used by the business middle platform and data middle platform to call basic component services. Business middle platform: Provides distributed scheduling and computation of models based on the model resource layer and data management layer, load balancing optimization of server resources, and efficient and stable communication and transmission of multi-source heterogeneous data, and is available for use by various business application components in the application layer; and Data platform: Used for data management during the process of calling model resource layer, and also covers the data resource publishing functions required for application layer visualization and interactive display, including: data processing and fusion, data sharing, data calculation and analysis, data storage; and is available for calling by various business application components in the application layer.

[0014] As an improvement to the above system, the data association mining module calls the state feature transformation analysis model and FP-Growth association rule mining model in the data association mining model in the model resource layer through the service scheduling layer to perform data association mining, and displays the results in the application layer and stores them in the data management layer. The knowledge reasoning and anomaly detection module calls the anomaly judgment template matching model and dynamic weighting model in the knowledge reasoning and anomaly detection model in the model resource layer through the service scheduling layer to perform knowledge reasoning and anomaly detection, and displays the results in the application layer and stores them in the data management layer.

[0015] On the other hand, the present invention also provides a method for detecting anomalies in the on-orbit status of spacecraft based on data association mining, comprising: Step 1: Data preprocessing. Perform corresponding data preprocessing and data cleaning according to the characteristics of different experimental tasks. Step 2: Data synchronization processing. Considering the characteristics of data generated by different channels in the aircraft, which are of multiple types, multiple sources, different frequencies, and different phases, data synchronization and phase adjustment are performed on the data. Step 3: Multi-source data fusion, which involves combining, analyzing, correlating, and fusing various types of information for the observations of the same variable; Step 4: Full-cycle data normalization. Perform full-cycle data normalization processing on the three types of data: structured, semi-structured, and unstructured, to convert the data values ​​to the same units and orders of magnitude. The data is merged at fixed intervals. Step 5: Data association mining. Based on the application characteristics of different experimental tasks, the relationships and patterns between various associated parameters and working states in the spacecraft telemetry parameters are obtained. The correlation degree is calculated and filtered to obtain rules and knowledge. Step 6: Knowledge Reasoning and Anomaly Detection. Based on the telemetry parameters and the sequence of spacecraft test actions, the obtained rules and knowledge are used to establish anomaly interpretation templates and rule bases to perform anomaly parameter detection. Telemetry parameter rules and spacecraft test mission status rules are formulated. The telemetry parameter rules are used to specify the value range of telemetry parameters in different test mission states of the spacecraft. The spacecraft test mission status rules are used to compare the action sequence of the spacecraft's on-orbit test mission to achieve spacecraft fault diagnosis.

[0016] Compared with existing technologies, the advantages of the spacecraft on-orbit state anomaly detection system and method based on data association mining of the present invention are as follows: 1. By receiving test data during the spacecraft mission, and based on the specific requirements and characteristics of the evaluation mission scenario, perform data cleaning work such as outlier removal, in-phase processing, and multi-source data fusion, and then perform full-cycle normalization processing on the cleaned data.

[0017] 2. It can perform unified retrieval, storage, management and sharing of multi-source heterogeneous data throughout the entire lifecycle, and features fast access speed and small storage capacity, and provides data visualization services based on the retrieval catalog.

[0018] 3. Based on the characteristics of different test missions such as track safety inspection, track capability repair, on-orbit operation, and fuel transfer and replenishment, a knowledge base and rule base for spacecraft test mission data are constructed. Data correlation mining and pattern analysis are conducted to provide technical support for fault diagnosis and abnormal state detection of spacecraft elements. Attached Figure Description

[0019] Figure 1 This is an architecture diagram of a spacecraft on-orbit state anomaly detection system and method based on data association mining. Figure 2 This describes the functional components of a spacecraft on-orbit state anomaly detection system and method based on data association mining. Figure 3 A flowchart illustrating the workflow of a spacecraft on-orbit state anomaly detection system and method based on data association mining; Figure 4 This is a schematic diagram of data processing with the same frequency and phase. Figure 5 This is a diagram illustrating data time synchronization. Figure 6 A schematic diagram for calculating the selection of ground stations; Figure 7 This is a schematic diagram of full-cycle data normalization; Figure 8 This is a structural diagram of the time pyramid model; Figure 9 Here is a flowchart of the FP-Growth algorithm; Figure 10 A diagram illustrating multi-source data classification, storage, and management methods; Figure 11 A line chart of the data before outlier removal; Figure 12 A line chart of the data after outlier removal; Figure 13 A line graph of the data before in-phase and in-frequency processing; Figure 14 This is a line graph of the data after in-phase and in-frequency processing; Figure 15 A graph showing the results of data normalization; Figure 16 This is a graph showing the results of the data fusion. Figure 17 A diagram illustrating how the dataset file path is stored in MySQL; Figure 18 This is a schematic diagram of the contents of the dataset for an in-orbit flight mission. Figure 19 A schematic diagram illustrating the frequent content collection for on-orbit flight missions; Figure 20 A schematic diagram showing the content of anomaly detection results for an on-orbit flight mission; Figure 21 This is a schematic diagram showing the results of abnormal state detection. Detailed Implementation

[0020] The technical solutions provided by the present invention will be further illustrated below with reference to the embodiments.

[0021] Example 1 This embodiment provides a spacecraft on-orbit status anomaly detection system based on data association mining, including: The architecture of the spacecraft on-orbit anomaly detection system and method based on data association mining is designed using the "platform + middleware + microservice" architecture concept. It adopts the design idea of ​​"basic support - data resources - data and business dual middleware services - typical applications". On the basis of basic components such as computing server clusters, storage servers, and other network devices, it provides rapid configuration and adaptive data processing services for multi-domain heterogeneous simulation resources for massive experimental data analysis and interactive display applications. It has the technical characteristics of openness, modularity, high performance, high reliability, reusability, assemblability, and secondary development capability, ensuring that the overall system is efficient, easy to use, stable, secure, and scalable.

[0022] The system can be divided into five layers from bottom to top: the infrastructure support layer, the data management layer, the model resource layer, the service scheduling layer, and the application layer. The infrastructure support layer and the network interconnection layer rely on existing hardware and software resources, such as... Figure 1 As shown.

[0023] (1) Basic support layer The basic support layer is the foundational support for the data sharing and analysis technology verification software. It mainly includes infrastructure such as computing server clusters, storage server clusters, and network equipment, providing a basic operating environment for the construction of this system.

[0024] (2) Data Management Layer The data management layer manages all types of data resources stored, transferred, and generated within the system. This primarily includes raw experimental data such as real-time / historical telemetry, telemetry, data transmission, and data transmission images from different aircraft channel sources, as well as simulated flight results from external ground simulation systems; data cleaning results after preprocessing such as data stitching, data interpolation or sampling, outlier removal, and format conversion; normalized results of the entire lifecycle of aircraft elements; prior rules and knowledge bases; results of pattern mining and knowledge reasoning analysis of multi-source experimental data; and results of fault diagnosis and abnormal state detection of aircraft elements.

[0025] Meanwhile, it combines large-scale spatial databases, database relational tables, and document catalogs based on mature commercial databases for storage, and adopts relational, non-relational, and time-series storage modes for different data types and application characteristics to store massive amounts of multi-source heterogeneous data.

[0026] (3) Model resource layer The model resource layer manages all data processing and computing models in the system in a unified manner. According to the model function, it is divided into data preprocessing model, data synchronous processing model, multi-source data fusion model, full-cycle data normalization model, data association mining model, and knowledge reasoning and anomaly detection model.

[0027] The data preprocessing models specifically include data acquisition, Lagrange interpolation models, and Dixon criterion models; the data synchronization models specifically include multi-channel data synchronization and multi-channel data time synchronization; the multi-source data fusion models specifically include weighted data fusion models; the full-cycle data normalization models specifically include time pyramid models; the data association mining models specifically include state feature transformation analysis and FP-Growth association rule mining models; and the knowledge reasoning and anomaly detection models specifically include anomaly interpretation template matching and dynamic weighted models.

[0028] (4) Service scheduling layer Basic components: These mainly include common basic components such as data access components, message middleware components, microservice development framework, API gateway, task scheduling framework, and log center. Both the business middle platform and the data middle platform can call these services.

[0029] Business middle platform: mainly includes large-scale distributed scheduling and computing of models based on model resource layer and data management layer, load balancing optimization of server resources, efficient and stable communication transmission of massive multi-source heterogeneous data, and calls to various business application components in the application layer.

[0030] Data platform: mainly includes data management during the process of calling model resource layer, and also covers data resource publishing functions such as data processing, data fusion, intelligent data push, data storage, and data sharing required for application layer visualization and interactive display, and is available for calling by various business application components in the application layer.

[0031] (5) Application layer The application layer includes display applications such as full-cycle data visualization, data correlation mining and analysis visualization, fault diagnosis and abnormal state monitoring, etc., providing an efficient, concise and intuitive human-computer interaction interface, and constructing different client application systems in the form of component-based service publishing and resource publishing. The implementation process of this invention is given in conjunction with the accompanying drawings to illustrate the technical solution of this invention in detail.

[0032] The module composition of the spacecraft on-orbit state anomaly detection system and method based on data association mining is as follows: Figure 2 As shown, it includes eight modules: data preprocessing module, data synchronous processing module, multi-source data fusion module, full-cycle data normalization module, data association mining module, knowledge reasoning and anomaly detection module, full-cycle data management and sharing module, and data analysis interactive display module.

[0033] The workflow of the spacecraft on-orbit state anomaly detection system and method based on data association mining is as follows: Figure 3 As shown, Figure 3 The table in the middle is for illustrative purposes only.

[0034] (1) Data preprocessing: First, test data from different channels of the aircraft are collected. The collected test data is preprocessed and duplicate or redundant data is deleted. When there are missing values ​​in the test data, the corresponding algorithm is called for interpolation. When there are outliers in the test data, the corresponding algorithm is called for outlier removal. Then the preprocessed test data is frequency-synchronized.

[0035] (2) Data in the same frequency and phase: In view of the characteristics of different channel data being multi-type, multi-source, different frequency and different phase, the experimental data from different channel sources are adjusted to be in the same frequency and phase, and these experimental data with different frequencies and different phases are processed into data under the same time scale and frequency, and then the subsequent multi-source data fusion processing operation is carried out.

[0036] (3) Multi-source data fusion: Through weighted data fusion algorithm, multiple telemetry parameters are optimized to achieve the integration, analysis, association and fusion of various information, so as to obtain more accurate, reliable and stable data content and improve the accuracy of subsequent mining results of multi-source data.

[0037] (4) Full-cycle data normalization: The data generated by multi-source fusion consists of data of different dimensions and orders of magnitude, which need to be normalized throughout the entire cycle. Dimensionless processing methods are used to reduce the gap between values. By constructing a time pyramid model, the experimental mission data throughout the entire lifecycle of the spacecraft are normalized and managed according to four levels: "day, week, month, and year," reducing the computational complexity when converting data of the same dimensions and orders of magnitude, and improving computational efficiency.

[0038] (5) Data association mining: The FP-Growth association rule mining algorithm is adopted. The iterative method of layer-by-layer search is used to find the relationship and pattern between the associated parameters and working status in the telemetry parameters of the aircraft. Through the calculation and screening of the association degree, rules and knowledge are finally formed, providing relevant rule support for the establishment of anomaly judgment templates and rule bases.

[0039] (6) Anomaly detection: Based on the anomaly interpretation rules obtained from data association mining and combined with expert prior knowledge, anomaly interpretation templates and rule bases are established to perform anomaly parameter detection, anomaly trend analysis, anomaly frequency analysis, anomaly interval analysis, and anomaly correlation analysis. This enables efficient statistical analysis of the changing trends of telemetry parameters over long time spans. At the same time, dynamic weighting of the anomaly interpretation templates is applied to address the complex transformation characteristics of telemetry data, effectively adapting to the changing characteristics of telemetry data from on-orbit spacecraft in complex space environments.

[0040] (7) Full-cycle data management and sharing: After extracting, cleaning, integrating, processing and summarizing the scattered metadata, a set of data products classified and graded for the full-cycle data of structured, semi-structured and unstructured aircraft test missions is formed. Advanced search functions such as fast data retrieval, complex condition query and statistical analysis are provided, and the data products can be shared among multiple users.

[0041] (8) Data analysis and interactive display: After the data cleaning, normalization, mining and other stages are completed, the final data processing results are generated and these calculation results are visualized. In automatic mode, the system will automatically receive data during the operation of the aircraft and analyze the data, displaying the full-cycle parameter data, the overall task process, data mining analysis results, etc. in the form of two-dimensional charts; in active mode, the system first receives the basic conditions selected by the user, such as the time period, task type, data category, number of parameters, etc., and then queries the data results based on these conditions, and then displays these data results in a two-dimensional format.

[0042] The main functions of each module are described below: (1) Data preprocessing module Based on the characteristics of different test missions such as track rescue, track obstacle removal, track transfer, and on-orbit refueling, the data preprocessing module adopts corresponding data preprocessing and data cleaning methods. It studies data preprocessing techniques such as Lagrange interpolation and Dixon discriminant analysis to achieve data cleaning work such as removing duplicate and redundant data, data splicing, data interpolation or sampling, outlier removal, and data format conversion. This enables data to be re-examined and verified, deleting duplicate information, correcting and identifying erroneous information in the data, and maintaining the correctness of the data.

[0043] (2) Data synchronous frequency and phase processing module The data synchronization and phase processing module addresses the characteristics of data generated from different channels in the aircraft, which are multi-type, multi-source, and of different frequencies and phases. It performs data synchronization and phase adjustment on test data from different channel sources and uses the synchronization information generation function to process these test data of different frequencies and phases into data at the same time scale and frequency for subsequent data fusion, normalization and other analysis and processing steps.

[0044] 1) Data synchronization processing: Based on a unified configuration of time intervals and data frequencies, the data is subtracted according to the time interval and frequency to achieve data synchronization. For example, within the T0-T10 time interval, after data is inserted, all data frequencies satisfy the same difference value. A schematic diagram of data synchronization processing is shown below. Figure 4 As shown.

[0045] 2) Multi-channel data time synchronization solves the problem of absolute time synchronization of data from different channels and with different results. Data processed to the same frequency has already resolved the issue of synchronization between different channels. Based on this, this submodule can align experimental data from different channels to a specific time point. However, due to inherent errors in the aircraft itself, as well as errors in the sampling data and the machine's local time, a unified timeline needs to be established to accurately display the experimental process. All received parameter information needs to be time-calibrated, unifying the corresponding parameter data to a suitable time point.

[0046] In data maintenance, two data buffers are established. The processed data buffer stores data according to a uniform time phase and interval to ensure data synchronization. On this basis, a time axis with different channel time synchronization is established (which may have a delay). Data is selected and packaged for transmission at a certain frequency.

[0047] First, a unified time axis T is established. The time difference is calculated based on the received telemetry parameters, considering errors inherent in the aircraft itself, errors in the sampling data, and errors in the machine's local time. Time calibration is then performed, unifying the corresponding parameters onto a single time axis T. Finally, the test data from different channels are aligned to a specific time point. A data time synchronization diagram is shown below. Figure 5 As shown.

[0048] (3) Multi-source data fusion module The multi-source data fusion module, utilizing algorithms such as weighted data fusion, can calculate a weighted arithmetic mean of observations of the same variable arranged chronologically, using the chronological order as weights. This mean is then used as a benchmark for optimizing multi-station telemetry data. Furthermore, it integrates, analyzes, correlates, and fuses various types of information to obtain more accurate, reliable, and stable data, thereby improving the accuracy of subsequent multi-source data mining results.

[0049] It supports user-defined fusion methods, allowing for customized fusion based on parameters such as telemetry type, data type (real-time, delayed), and ground station ratio. The default fusion method uses a real-time data fusion algorithm, which calculates the distance between the ground station and the satellite in real time and selects the best data from multiple telemetry sources based on the principle of prioritizing nearby stations and ensuring that the delayed data is greater than the real-time data.

[0050] The formula for calculating satellite distance is:

[0051]

[0052]

[0053] The specific methods for calculating and selecting ground stations are as follows: Figure 6 As shown.

[0054] (4) Full-cycle data normalization module The processes of spacecraft long-range approach, orbital rendezvous, close approach, hovering, and evacuation, as well as on-orbit maintenance, attitude and orbit takeover, and space operations, generate a large amount of experimental data. This data, with its different dimensions and orders of magnitude, needs to be normalized to convert its values ​​to the same dimensions and orders of magnitude. The full-cycle data normalization module performs full-cycle data normalization on the massive amounts of pre-processed and fused analysis data from spacecraft experimental data, based on three data types: structured, semi-structured, and unstructured. Figure 7 As shown, Figure 7 The table in the middle is for illustrative purposes only. It transforms the data values ​​to the same units and orders of magnitude, forming normalization processing results and data products of different dimensions such as hierarchical, regional, and block-based.

[0055] The concept of a time pyramid model is proposed to reduce computational complexity and improve efficiency when converting data of the same dimensions and orders of magnitude. In this model, experimental mission data throughout the entire spacecraft lifecycle is normalized and managed according to four levels: daily, weekly, monthly, and yearly. This allows experimental mission data to be merged at fixed times each day, week, month, and year. The specific hierarchical relationships are as follows: Figure 8 As shown.

[0056] (5) Data Association Mining Module The data association mining module is designed for different experimental tasks. It adopts the FP-Growth association rule mining algorithm and uses an iterative method of layer-by-layer search to find the relationships and patterns between various associated parameters and working states in the spacecraft telemetry parameters. Finally, it forms rules and knowledge, providing relevant rule support for the establishment of anomaly judgment templates and rule bases.

[0057] The FP-Growth association rule mining algorithm is a frequent pattern growth algorithm that directly generates frequent sets without producing candidate sets, effectively generating association rules. After the first scan, this method compresses the frequent itemsets in the database into a frequent pattern tree (FP-Tree) while retaining the association information. Then, the FP-Tree is differentiated into several condition libraries, each associated with a frequent itemset of length L, and then mining is performed on these condition libraries separately. The process of constructing the frequent pattern tree using FP-Growth is as follows: Figure 9 As shown.

[0058] (1) Scan the database once to generate L-frequent sets, sort them in descending order, and put them into the L table; (2) Create a root node and mark it as null. Scan the database once. When you get a set of items in the database, arrange the elements in it according to the order of the L table. Then recursively call FP-Growth to implement FP-Tree growth.

[0059] After obtaining the FP-tree, association rules can be generated. First, working backward from the end of a branch, if an item meets the minimum support requirement, then the frequency of that item in other branches is counted and the confidence level is calculated. If the minimum confidence level is not met, the rule is discarded. As can be seen from the FP-tree, each branch contains multiple items, so frequent multisets can be found from the end to the beginning, similar to finding frequent singlets.

[0060] The telemetry parameters mined include two types: discrete state variables and continuously changing analog variables. State variables themselves are finite sets, so they do not require piecewise feature representation. However, they need to be processed to construct a transaction set suitable for the FP-Growth algorithm. Since state values ​​are not unique, they cannot be directly used for mining. Therefore, the parameter name and state value are combined into one item to ensure the uniqueness of the item. An example of constructing a state variable transaction set is shown in Table 1.

[0061] Table 1. Construction Table of State Quantity Transaction Set

[0062] For analog time series, the multidimensional time series is segmented and symbolically represented based on trends. For example, it is divided into time periods such as 1, 2, 3, etc., and then symbolically represented according to the data change trend, such as U (up) for data increase and D (down) for data decrease. Finally, each parameter consists of a set of symbols representing trend characteristics. The set of symbolic feature subsequences of multidimensional parameters at the same time moment is considered as a transaction. The set of multidimensional time series constitutes the transaction set required for mining, as shown in Table 2.

[0063] Table 2. Analog Transaction Set Construction Table

[0064] Mining is performed based on the constructed transaction set. The validation metrics are defined as follows: Accuracy = (Number of known rules mined M_known_count / Total number of mining attempts M_all_count) * 100%; Redundancy = (Number of all rules mined M_all_once - Number of known correct rules mined M_right_once) / Number of all rules mined M_all_once * 100% (related to the confidence threshold setting). Higher accuracy indicates better mining results. Redundancy represents the proportion of known rules mined. Ideally, redundancy should be as low as possible. Redundancy is related to the confidence setting; a higher confidence setting results in fewer rules mined, potentially lowering redundancy, but it may also lead to the loss of known correct rules. Furthermore, excessively high confidence settings may result in the loss of previously unknown potential association rules. Therefore, the confidence setting needs to be within a reasonable range. Finally, the mined association rules are uniformly stored and managed.

[0065] (6) Knowledge Reasoning and Anomaly Detection Module The knowledge reasoning and anomaly detection module comprehensively utilizes the clustering features, association patterns, and prior knowledge provided by the data association mining module, as well as the association mining results. By establishing anomaly interpretation templates and rule bases, it can efficiently statistically analyze the changing trends of telemetry parameters over long time spans.

[0066] Based on the telemetry parameter rules and the aircraft test mission status rules, and combined with the discovered correlation patterns, anomaly interpretation templates and rule bases were established. The telemetry parameter rules are mainly used to specify the value range of different telemetry parameters in different test mission statuses of the aircraft. The table structure is shown in Table 3.

[0067] Table 3 Structure of Telemetry Parameter Rule Table

[0068] The aircraft test mission plan table is used to record the sequence of actions for on-orbit test missions of the aircraft. It mainly includes the name of the working state, the start time of the action, and the end time of the action. The table structure is shown in Table 4.

[0069] Table 4. Structure of the Aircraft Work Plan

[0070] Anomaly detection and diagnosis based on a knowledge base is achieved by constructing keyword functions. Keyword functions are the smallest units that implement the specific content of rules. By parsing the rule content (including the correspondence between instructions and parameters, and the parameter threshold range before and after the instruction is triggered), the rule content and keyword functions are matched, thereby achieving automatic detection. Based on existing inference rules, the preprocessed parameter values ​​are judged according to the normal threshold range to determine whether the parameters are in an abnormal stage. Further inference based on the abnormal conditions leads to the fault and its location.

[0071] (7) Full-cycle data management and sharing module Through preliminary multi-source heterogeneous data cleaning, intelligent data mining and pattern analysis, and full lifecycle data normalization, the scattered metadata is extracted, cleaned, integrated, processed, and summarized. The full lifecycle data management and sharing module then handles structured, semi-structured, and unstructured data from the entire lifecycle of aircraft test missions, such as... Figure 10 As shown, this system enables high-bandwidth storage and reading / writing of multi-source heterogeneous data, forming a collection of data products categorized and graded according to experimental data. It also provides advanced search functions such as fast data retrieval, complex conditional queries, and statistical analysis. Furthermore, it implements role-based access control for experimental data, limiting the scope of accessible products based on different roles, and provides three methods—interface, file, and database retrieval—to support the sharing of relevant data products among multiple users.

[0072] (8) Data analysis interactive display module The data analysis and interactive display module primarily implements data analysis and display functions such as full-cycle data display, data mining result display, and knowledge reasoning and pattern analysis result display, as well as interactive operation functions such as data management and sharing, and manual data processing. The module has two usage modes: In automatic mode, it automatically receives data from the spacecraft's operation and analyzes it, displaying full-cycle parameter data, the overall mission flow, and data mining analysis results in two-dimensional charts; in active mode, it first receives basic conditions selected by the user, such as the time period, mission type, data category, and number of parameters, then retrieves the data results based on these conditions and displays these results in a two-dimensional format.

[0073] Simulation Examples (1) Outlier removal When querying a certain telemetry parameter, a sudden increase in value was found at 00:02:50 on 2023 / 01 / 02, which was determined to be an outlier. Figure 11 As shown.

[0074] After identifying outliers, outlier removal was performed on the parameter. Following outlier removal, the spikes disappeared, and the data returned to a stable state with no sudden increases, decreases, or out-of-range data. The results are as follows: Figure 12 As shown.

[0075] (2) Same frequency and phase processing function After removing outliers from a certain telemetry parameter, it was found that the data had issues such as different frequencies, different phases, and different data frequencies. Figure 13 As shown.

[0076] After identifying the problem, the parameter was processed to be synchronized with the same frequency and phase. A step size of 1 second was set, with one frame of data per second. This synchronized processing transformed the data into data with the same time scale and frequency, which was then used for subsequent data fusion, normalization, and other analytical processing. The processing results are as follows: Figure 14 As shown.

[0077] (3) Data normalization function Since the basic unit of the time pyramid model is the day, and considering the large number of telemetry parameters, secondary sampling and storage of telemetry parameters are performed on a daily basis. In data playback mode, all telemetry parameter data at 0:0:0:000 for each day are stored in a database table. A query displays the normalized result of a specific telemetry parameter from 2023 / 01 / 02 to 2023 / 01 / 25. The results are as follows. Figure 15 As shown.

[0078] (4) Multi-source data fusion experiment Multi-source data fusion integrates data from different channels, including real-time and delayed data, and data from different ground stations. The fusion results of multi-channel telemetry parameters are shown in the figure. Figure 16 As shown.

[0079] (5) Data mining and anomaly detection Collect spacecraft position and velocity data during on-orbit missions, fuse the position and velocity data to construct a dataset, and store it in MySQL as a file path, such as... Figure 17 As shown, Figure 17 For illustrative purposes only.

[0080] Based on the on-orbit flight mission dataset, a transaction set suitable for the FP-Growth algorithm is constructed. Telemetry parameters include two types: discrete state variables and continuously changing analog variables. For discrete state variables, "parameter name + state value" is combined into one item to ensure uniqueness. For continuous analog variables, the multidimensional time series is symbolized as a trend (U increases / D decreases), and the transaction set is constructed using the symbolized subsequences of each parameter at the same time point, such as... Figure 18 As shown.

[0081] Mining datasets generated from on-orbit flight missions can produce frequent sets for anomaly rule analysis. The FP-Growth association rule mining algorithm, a frequent pattern growth algorithm that directly generates frequent sets without producing candidate sets, effectively produces association rules. After the first scan, this method compresses the frequent item sets in the database into a frequent pattern tree (FP-Tree) while retaining the association information. The FP-Tree is then divided into condition libraries, each associated with a frequent item set of length L, and mining is performed on these condition libraries separately. Finally, based on the relationships and patterns between associated parameters and operational states in the spacecraft telemetry parameters, rules and knowledge are formed through association degree calculation and filtering. This provides relevant rule support for the establishment of anomaly interpretation templates and rule bases. The generated frequent sets are as follows: Figure 19 As shown, Figure 19 For illustrative purposes only.

[0082] Based on the data analysis results from the on-orbit flight mission, anomaly analysis is performed on the subsequent experimental data generated by the mission. This allows for the detection of anomalous parameters and other information, enabling the transmission of appropriate anomaly alerts to the mission operators. The results of anomaly detection during the on-orbit flight mission are as follows: Figure 20 , Figure 21 As shown, Figure 20 For illustrative purposes only.

[0083] Example 2 This embodiment provides a method for detecting anomalies in the on-orbit status of spacecraft based on data association mining, including: Step 1: Data preprocessing. Perform corresponding data preprocessing and data cleaning according to the characteristics of different experimental tasks. Step 2: Data synchronization processing. Considering the characteristics of data generated by different channels in the aircraft, which are of multiple types, multiple sources, different frequencies, and different phases, data synchronization and phase adjustment are performed on the data. Step 3: Multi-source data fusion, which involves combining, analyzing, correlating, and fusing various types of information for the observations of the same variable; Step 4: Full-cycle data normalization. Perform full-cycle data normalization processing on the three types of data: structured, semi-structured, and unstructured, to convert the data values ​​to the same units and orders of magnitude. The data is merged in fixed cycles of days, weeks, months, and years. Step 5: Data association mining. Based on the application characteristics of different experimental tasks, the relationships and patterns between various associated parameters and working states in the spacecraft telemetry parameters are obtained. The correlation degree is calculated and filtered to obtain rules and knowledge. Step 6: Knowledge Reasoning and Anomaly Detection. Based on the telemetry parameters and the sequence of spacecraft test actions, the obtained rules and knowledge are used to establish anomaly interpretation templates and rule bases to perform anomaly parameter detection. Telemetry parameter rules and spacecraft test mission status rules are formulated. The telemetry parameter rules are used to specify the value range of telemetry parameters in different test mission states of the spacecraft, and the spacecraft test mission status rules are used to compare the action sequence of the spacecraft's on-orbit test mission.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A spacecraft on-orbit state anomaly detection system based on data association mining, characterized in that, include: Data preprocessing module: used to perform corresponding data preprocessing and data cleaning according to the characteristics of different experimental tasks; Data frequency and phase processing module: used to perform data frequency and phase adjustment based on the different types, sources, frequencies and phases of data generated by different channels in the aircraft; Multi-source data fusion module: used to integrate, analyze, correlate, and fuse multiple types of information for observations of the same variable; Full-cycle data normalization module: Used to perform full-cycle data normalization processing for three types of data: structured, semi-structured, and unstructured, so as to convert data values ​​to the same units and orders of magnitude; Data association mining module: Based on the application characteristics of different experimental tasks, it targets discrete state quantities and continuous analog quantities in telemetry parameters. Discrete state quantities are represented by a combination of parameter names and state values, while continuous analog quantities are represented by trend symbols. Then, a transaction set is uniformly constructed, and the rules and knowledge between various associated parameters and working states in the spacecraft telemetry parameters are mined through the construction of a frequent pattern tree. and The knowledge reasoning and anomaly detection module is used to establish anomaly judgment templates and rule bases based on the rules and knowledge obtained from the data association mining module for telemetry parameters and spacecraft test action sequences. It also formulates telemetry parameter rules and spacecraft test mission status rules. The telemetry parameter rules are used to specify the value range of telemetry parameters in different test mission states of the spacecraft, and the spacecraft test mission status rules are used to compare the action sequences of the spacecraft's on-orbit test missions to achieve spacecraft fault diagnosis.

2. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 1, characterized in that, Also includes: Full-cycle data management and sharing module: used for high-bandwidth storage, read / write, and role-based access control of multi-source heterogeneous data for the entire lifecycle of aircraft test missions; and Data Analysis Interactive Display Module: Used to perform analysis and display on full-cycle data, data mining results, knowledge reasoning and pattern analysis results, as well as interactive operations including data management and sharing, and manual data processing.

3. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 1, characterized in that, The data association mining module uses the FP-Growth association rule mining algorithm, and the specific execution steps are as follows: Step 1: Construct a transaction set using normalized aircraft telemetry parameters; Step 2: Scan the transaction set, generate frequent item sets and sort them in descending order of support, then store them in table L; Step 3: Create the root node of the FP tree, and recursively insert the items in each transaction according to the order of the L table to build a frequent pattern tree; Step 4: Mine frequent itemsets. Starting from the end of each branch of the FP-tree and working backwards, count the items that meet the minimum support and their frequency of occurrence. Step 5: Set minimum support and confidence thresholds, filter rules that meet the confidence level, and balance the redundancy rate of rule mining with the retention rate of known rules; Step 6: Evaluate the mining accuracy and redundancy rate, and adjust the confidence interval to retain potential unknown rules.

4. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 1, characterized in that, The knowledge reasoning and anomaly detection module performs the following steps: Step 1: Match the aircraft telemetry parameters with the anomaly interpretation template; Step 2: Perform dynamic weighting of spacecraft telemetry parameter templates; Step 3: Conduct anomaly detection and assessment of aircraft telemetry parameters; Step 4: If the evaluation is abnormal, proceed to step 5; otherwise, proceed to step 3 for the next spacecraft telemetry parameter until the traversal is complete. Step 5: Record abnormal states of the aircraft's telemetry parameters.

5. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 1, characterized in that, The system architecture is divided into 5 layers, including: Basic support layer: used for data sharing and analysis technology verification software, including computing server clusters, storage server clusters, and network devices; Data management layer: Used to manage various data resources that are stored, transferred, and generated, and to store multi-source heterogeneous data; Model resource layer: used for unified management of various data processing and computing models; Service scheduling layer: used to provide services for basic components, business middle platform, and data middle platform; and Application layer: Used to display applications, provide human-computer interaction interfaces, and form different application clients.

6. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 5, characterized in that, The data management layer manages various data resources, including: real-time and historical telemetry, telemetry, data transmission, data transmission images, and simulated flight results from external ground simulation systems, as well as raw experimental data; data cleaning results after preprocessing such as data splicing, data interpolation or sampling, outlier removal, and format conversion; normalized results of the entire lifecycle of aircraft elements; prior rules and knowledge base data; results of pattern mining and knowledge reasoning analysis of multi-source experimental data; and results of fault diagnosis and abnormal state detection of aircraft elements. The data is stored in combination with a large-scale spatial database, database relational tables, and document cataloging library based on commercial databases. Relational, non-relational, and time-series storage modes are adopted for different data types and application characteristics to store multi-source heterogeneous data.

7. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 5, characterized in that, The data processing and computing model under the unified management of the model resource layer includes: Data preprocessing models include data extraction, Lagrange interpolation model, and Dixon criterion model; Data synchronous frequency and phase processing model: including multi-channel data synchronous frequency and multi-channel data time synchronization; Multi-source data fusion models: including weighted data fusion models; Full-cycle data normalization model: including the time pyramid model; Data association mining models include state feature transformation analysis and FP-Growth association rule mining models; and Knowledge reasoning and anomaly detection models: including anomaly interpretation template matching and dynamic weighted models.

8. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 5, characterized in that, The service scheduling layer specifically includes: Basic components: These include common basic components such as data access components, message middleware components, microservice development framework, API gateway, task scheduling framework, and log center, which are used by the business middle platform and data middle platform to call basic component services. Business middle platform: Provides distributed scheduling and computation of models based on the model resource layer and data management layer, load balancing optimization of server resources, and efficient and stable communication and transmission of multi-source heterogeneous data, and is available for use by various business application components in the application layer; and Data platform: Used for data management during the process of calling model resource layer, and also covers the data resource publishing functions required for application layer visualization and interactive display, including: data processing and fusion, data sharing, data calculation and analysis, data storage; and is available for calling by various business application components in the application layer.

9. The spacecraft on-orbit status anomaly detection system based on data association mining as described in claim 5, characterized in that, The data association mining module calls the state feature transformation analysis model and FP-Growth association rule mining model in the data association mining model in the model resource layer through the service scheduling layer to perform data association mining, and displays the results in the application layer and stores them in the data management layer. The knowledge reasoning and anomaly detection module calls the anomaly judgment template matching model and dynamic weighting model in the knowledge reasoning and anomaly detection model in the model resource layer through the service scheduling layer to perform knowledge reasoning and anomaly detection, and displays the results in the application layer and stores them in the data management layer.

10. A method for detecting anomalies in the on-orbit status of spacecraft based on data association mining, comprising: Step 1: Data preprocessing. Perform corresponding data preprocessing and data cleaning according to the characteristics of different experimental tasks. Step 2: Data synchronization processing. Considering the characteristics of data generated by different channels in the aircraft, which are of multiple types, multiple sources, different frequencies, and different phases, data synchronization and phase adjustment are performed on the data. Step 3: Multi-source data fusion, which involves combining, analyzing, correlating, and fusing various types of information for the observations of the same variable; Step 4: Full-cycle data normalization. Perform full-cycle data normalization processing on the three types of data: structured, semi-structured, and unstructured, to convert the data values ​​to the same units and orders of magnitude. The data is merged at fixed intervals. Step 5: Data association mining. Based on the application characteristics of different experimental tasks, the relationships and patterns between various associated parameters and working states in the spacecraft telemetry parameters are obtained. The correlation degree is calculated and filtered to obtain rules and knowledge. Step 6: Knowledge Reasoning and Anomaly Detection. Based on the telemetry parameters and the sequence of spacecraft test actions, the obtained rules and knowledge are used to establish anomaly interpretation templates and rule bases to perform anomaly parameter detection. Telemetry parameter rules and spacecraft test mission status rules are formulated. The telemetry parameter rules are used to specify the value range of telemetry parameters in different test mission states of the spacecraft. The spacecraft test mission status rules are used to compare the action sequence of the spacecraft's on-orbit test mission to achieve spacecraft fault diagnosis.

Images