Method, system and storage medium for integrity verification and correction of flight data

Abstract

The present application relates to the technical field of data verification, and discloses a method, system and storage medium for integrity verification and correction of flight data. The method includes: performing spatiotemporal feature extraction processing on a multimodal feature data stream, constructing an association matrix according to a multi-level data mapping relationship, and obtaining a spatiotemporal feature association network; performing hierarchical verification processing on the spatiotemporal feature association network to obtain abnormal marked data; performing regional recursive positioning processing on the abnormal marked data to obtain an abnormal propagation link diagram; performing correction scheme design processing on the abnormal propagation link diagram, and obtaining a data correction execution sequence based on data reconstruction and impact assessment optimization schemes; performing progressive processing on the data correction execution sequence to obtain corrected flight data. The present application establishes a data integrity verification and correction mechanism in a distributed data acquisition environment, which can not only accurately identify and locate data anomalies, but also ensure that the corrected data meets physical constraints and spatiotemporal consistency requirements.

Description

Technical Field

[0001] The present application relates to the field of data verification, and in particular to a method, system and storage medium for integrity verification and correction of flight data. Background Art

[0002] At present, the flight data recording system in the civil aviation field mainly uses centralized data collection and fixed threshold detection to perform data integrity verification. Traditional flight data recorders collect flight parameters through a single data bus and perform batch processing and analysis on the ground. This method is susceptible to single point failures during the data collection process, and the data processing lacks real-time performance. At the same time, the existing data integrity verification method mainly relies on a preset fixed threshold for anomaly detection. The inspection rules are relatively simple and it is difficult to deal with complex data anomalies. In terms of data correction, the existing technology usually uses a simple interpolation or average value replacement method, which lacks in-depth analysis and consideration of the correlation between data.

[0003] However, these traditional methods have obvious shortcomings: first, the centralized data collection architecture has low reliability. Once the core collection node fails, the entire system will be paralyzed; second, the fixed threshold inspection method cannot adapt to the dynamic changes during the flight process and is prone to false alarms or omissions; third, the simple data correction method does not consider the temporal and spatial correlation between the data. The corrected data may violate physical constraints and affect the availability of the data; finally, there is a lack of a systematic data quality assurance mechanism, making it difficult to ensure the integrity and consistency of the corrected data. Summary of the invention

[0004] The present application provides a method, system and storage medium for integrity verification and correction of flight data, which are used to establish an adaptive data integrity verification and correction mechanism in a distributed data acquisition environment, which can not only accurately identify and locate data anomalies, but also ensure that the corrected data meets physical constraints and spatiotemporal consistency requirements.

[0005] In a first aspect, the present application provides a method for integrity verification and correction of flight data, which includes: performing distributed collection and processing on flight data, filtering data through a dynamic weight self-adjustment mechanism to obtain a multimodal feature data stream; performing spatiotemporal feature extraction processing on the multimodal feature data stream, constructing an association matrix according to a multi-level data mapping relationship, and obtaining a spatiotemporal feature association network; performing hierarchical inspection processing on the spatiotemporal feature association network, performing data verification according to a three-level architecture of rapid warning, pattern recognition, and semantic verification to obtain abnormal marked data; performing regional recursive positioning processing on the abnormal marked data, combining multi-level threshold filtering with regional diffusion tracking to obtain an abnormal propagation link diagram; performing correction scheme design processing on the abnormal propagation link diagram, and obtaining a data correction execution sequence based on data reconstruction and impact assessment optimization schemes; performing progressive processing on the data correction execution sequence, using a segmented state check and a dynamic rollback mechanism to correct the data, and obtaining corrected flight data.

[0006] In a second aspect, the present application provides a system for integrity checking and correction of flight data, the system for integrity checking and correction of flight data comprising:

[0007] The acquisition module is used to perform distributed acquisition and processing of flight data, filter data through a dynamic weight self-adjustment mechanism, and obtain a multi-modal feature data stream;

[0008] The extraction module is used to perform spatiotemporal feature extraction processing on the multimodal feature data stream, construct an association matrix according to the multi-level data mapping relationship, and obtain a spatiotemporal feature association network;

[0009] The verification module is used to perform hierarchical verification processing on the spatiotemporal feature association network, perform data verification according to the three-level architecture of rapid warning, pattern recognition, and semantic verification, and obtain abnormal labeled data;

[0010] The positioning module is used to implement regional recursive positioning processing on the abnormal marked data, combining multi-level threshold filtering and regional diffusion tracking to obtain the abnormal propagation link map;

[0011] A processing module is used to design and process a correction scheme for the abnormal propagation link diagram, and obtain a data correction execution sequence based on data reconstruction and impact assessment optimization scheme;

[0012] The correction module is used to implement progressive processing on the data correction execution sequence, use segmented status checking and dynamic rollback mechanism to correct the data, and obtain corrected flight data.

[0013] A third aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned integrity check and correction method for flight data.

[0014] In the technical solution provided by the present application, the risk of single-point collection failure is avoided and the reliability and robustness of data collection are improved by performing distributed collection and processing of flight data and introducing a dynamic weight self-adjustment mechanism; spatiotemporal feature extraction is performed on multimodal feature data streams and a multi-level data mapping relationship is constructed, which realizes the comprehensive capture and in-depth analysis of data features and provides a reliable feature basis for anomaly detection; a three-level architecture of rapid warning, pattern recognition and semantic verification is adopted for hierarchical inspection, and a multi-level anomaly detection defense line is constructed, which significantly improves the accuracy and reliability of anomaly detection; the spatial distribution characteristics of abnormal data are accurately identified by combining regional recursive positioning and multi-level threshold filtering, and the propagation law and impact range of anomalies are accurately grasped by combining the regional diffusion tracking mechanism; in the design of the correction scheme, based on the dual guarantee of data reconstruction and impact assessment, the scientificity and feasibility of the correction scheme are ensured, and the negative impact of the correction process on related data is effectively reduced; in the process of correction implementation, a progressive processing strategy is adopted, problems are discovered in time through segmented status inspection, and errors are corrected in time by relying on a dynamic rollback mechanism, which greatly improves the reliability and stability of data correction. The overall solution forms a complete data processing closed loop, with each processing link closely connected and mutually supporting, which not only ensures the integrity and accuracy of flight data, but also provides a more reliable data basis for flight safety analysis. Especially in terms of anomaly detection and data correction, through multi-level verification and progressive correction strategies, it not only ensures the accuracy of processing, but also ensures the controllability of the correction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 A schematic diagram of an embodiment of a method for integrity checking and correction of flight data in an embodiment of the present application;

[0017] Figure 2 Schematic diagram of an embodiment of a system for integrity checking and correction of flight data in an embodiment of the present application. DETAILED DESCRIPTION

[0018] Embodiments of the present application provide a method, system and storage medium for integrity verification and correction of flight data. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the method for integrity checking and correction of flight data in the embodiment of the present application includes:

[0020] Step S101, performing distributed collection and processing on the flight data, filtering the data through a dynamic weight self-adjustment mechanism, and obtaining a multi-modal feature data stream;

[0021] Step S102: performing spatiotemporal feature extraction processing on the multimodal feature data stream, constructing an association matrix according to the multi-level data mapping relationship, and obtaining a spatiotemporal feature association network;

[0022] Step S103, performing hierarchical verification processing on the spatiotemporal feature association network, performing data verification according to the three-level architecture of rapid early warning, pattern recognition, and semantic verification, and obtaining abnormal labeled data;

[0023] Step S104: performing regional recursive positioning processing on the abnormal marked data, combining multi-level threshold filtering and regional diffusion tracking to obtain an abnormal propagation link diagram;

[0024] Step S105: Design a correction scheme for the abnormal propagation link diagram, and obtain a data correction execution sequence based on data reconstruction and impact assessment optimization scheme;

[0025] Step S106: progressively process the data correction execution sequence, use segmented status checking and dynamic rollback mechanism to correct the data, and obtain corrected flight data.

[0026] It is understandable that the execution subject of the present application may be a system for integrity checking and correction of flight data, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0027] Optionally, the flight data is collected and preliminarily processed through distributed collection and processing. Flight data refers to various types of data generated by the aircraft during flight, including information such as altitude, speed, attitude, and engine parameters. Distributed collection is to arrange multiple data collection nodes at key parts of the aircraft, each node independently collects and preprocesses data, and screens the data through a dynamic weight self-adjustment mechanism. The dynamic weight self-adjustment mechanism dynamically calculates the weight value based on factors such as the credibility of the data and the collection time, and conducts a comprehensive evaluation of the collected data. For example, when a sensor data fluctuates violently, the weight value of the sensor will be reduced accordingly, thereby reducing the impact of abnormal data on the overall data stream. The multimodal feature data stream obtained after such processing contains the temporal features and spatial feature information of multiple data sources. The multimodal feature data stream is subjected to spatiotemporal feature extraction processing. This process first analyzes the characteristics of the data in the time dimension, including the periodicity, trend, and mutation of the data, and considers the distribution characteristics of the data in the spatial dimension. Multi-level data mapping relationship refers to the establishment of association relationships between data at different levels, from micro data point association to macro data group association. The correlation matrix constructed in this way reflects the intrinsic connection between the data and ultimately forms a spatiotemporal feature correlation network.

[0028] When performing hierarchical inspection on the spatiotemporal feature association network, a three-level architecture is used for comprehensive inspection. The rapid warning stage mainly detects the mutation of the data and identifies abnormal fluctuations by setting dynamic thresholds; the pattern recognition stage conducts in-depth analysis of the data to identify abnormal patterns; the semantic verification stage verifies the rationality of the data from the perspective of business rules. These three levels are progressive, ensuring the accuracy and comprehensiveness of anomaly detection. The regional recursive positioning processing stage accurately locates the detected abnormal data. First, the initial impact area of ​​the abnormal data is determined, and then the positioning range is gradually narrowed by recursion. The multi-level threshold filtering mechanism filters abnormal data step by step by setting thresholds at different levels, while the regional diffusion tracking analyzes the propagation path and impact range of the anomaly, and finally generates an abnormal propagation link diagram.

[0029] When designing a correction scheme based on the abnormal propagation link diagram, it is necessary to consider the data integrity requirements and physical constraints. First, extract the key nodes, analyze the data missing situation, and then reconstruct the data, while evaluating the impact range of the correction operation. This process needs to balance the correction effect and the impact on related data, and finally generate a data correction execution sequence. The final progressive processing stage is to implement the data correction execution sequence in steps. The segmented status check mechanism checks each key node to ensure the correctness of the correction process, and the dynamic rollback mechanism rolls back to a safe state in time when a problem is found. In this way, the reliability and stability of data correction are guaranteed.

[0030] For example, during a flight, an aircraft's altitude sensor was abnormal. The distributed acquisition node collected data from the sensor that fluctuated dramatically within 10 seconds, jumping suddenly from a cruising altitude of 10,000 meters to 12,000 meters, while data from other related sensors such as pressure altitude and radar altitude remained stable. The dynamic weight self-adjustment mechanism immediately reduced the weight of the sensor from the standard weight of 1.0 to 0.3. The spatiotemporal feature extraction process found that this anomaly did not match the surrounding sensor data, and the rapid warning mechanism triggered an alarm. Pattern recognition confirmed that this mutation pattern was a typical sensor failure feature, and semantic verification verified that such a large altitude change could not occur during the level flight phase. Regional recursive positioning determined that the anomaly was limited to the sensor data and did not affect other systems. In the correction scheme design phase, a correction sequence was generated based on the values ​​of other altitude data sources combined with the aircraft's kinematic model. Finally, through progressive processing, the abnormal data segment was corrected to a value that conforms to physical laws. The corrected altitude data showed that the aircraft had been maintaining a cruising altitude of 10,000 meters. The entire correction process was a smooth transition without introducing new data anomalies. This example demonstrates the effectiveness and accuracy of the approach in dealing with anomalies in real flight data.

[0031] In the embodiment of the present application, by performing distributed collection and processing of flight data and introducing a dynamic weight self-adjustment mechanism, the risk of single-point collection failure is avoided, and the reliability and robustness of data collection are improved; spatiotemporal feature extraction is performed on multimodal feature data streams and a multi-level data mapping relationship is constructed, which realizes the comprehensive capture and in-depth analysis of data features and provides a reliable feature basis for anomaly detection; a three-level architecture of rapid warning, pattern recognition and semantic verification is adopted for hierarchical inspection, and a multi-level anomaly detection defense line is constructed, which significantly improves the accuracy and reliability of anomaly detection; the spatial distribution characteristics of abnormal data are accurately identified by combining regional recursive positioning and multi-level threshold filtering, and the propagation law and impact range of anomalies are accurately grasped by combining the regional diffusion tracking mechanism; in the design of the correction scheme, based on the dual guarantee of data reconstruction and impact assessment, the scientificity and feasibility of the correction scheme are ensured, and the negative impact of the correction process on related data is effectively reduced; in the process of correction implementation, a progressive processing strategy is adopted, problems are discovered in time through segmented status inspection, and errors are corrected in time by relying on a dynamic rollback mechanism, which greatly improves the reliability and stability of data correction. The overall solution forms a complete data processing closed loop, with each processing link closely connected and mutually supporting, which not only ensures the integrity and accuracy of flight data, but also provides a more reliable data basis for flight safety analysis. Especially in terms of anomaly detection and data correction, through multi-level verification and progressive correction strategies, it not only ensures the accuracy of processing, but also ensures the controllability of the correction process.

[0032] In an optional embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) Perform time segmentation processing on the flight data, split the data according to the characteristics of the flight phase, obtain the initial time series data set, and mark the outliers on the initial time series data set. Set the threshold boundary according to the data distribution characteristics to obtain the marked time series data;

[0034] (2) Perform weight calculation on the marked time series data, calculate the weight coefficient according to the data credibility index, obtain the initial weight matrix, and dynamically update the initial weight matrix, adjust the weight value according to the data fluctuation state, and obtain the target weight matrix;

[0035] (3) Perform feature decomposition on the target weight matrix, extract key features based on principal component analysis of the data, obtain a feature vector set, and perform dimensionality reduction on the feature vector set. Select feature dimensions based on the information retention rate to obtain a reduced-dimensional feature matrix.

[0036] (4) Clustering the reduced-dimensional feature matrix, dividing the feature groups according to data similarity, obtaining the feature clustering results, and optimizing the boundaries of the feature clustering results, adjusting the clustering boundaries according to the inter-class distance, and obtaining the candidate feature categories;

[0037] (5) Perform multimodal fusion processing on the candidate feature categories, build fusion rules based on the correlation between feature categories, and obtain a multimodal feature data stream.

[0038] Optionally, time segmentation is performed, which divides the continuous flight data according to the flight phase, including takeoff, climb, cruise, descent and landing. The basis for data segmentation is the characteristic parameters of each phase. For example, the characteristics of the takeoff phase are a rapid increase in ground speed and the beginning of the altitude climb, while the cruise phase is characterized by relatively stable altitude and speed. When marking the outliers of the initial time series data set after segmentation, the mean and standard deviation of the data are calculated using statistical analysis methods, and the threshold boundary is set to the mean plus or minus three times the standard deviation. Data points beyond this boundary are marked as potential outliers. The marked time series data enters the weight calculation stage, and the data credibility indicators include multiple dimensions such as data volatility, continuity of the acquisition timestamp, and consistency with related sensor data. The weight coefficient is calculated using a multi-factor weighted average method, and the initial weight of each indicator is set according to empirical values, such as a volatility weight of 0.4, a time continuity weight of 0.3, and a data consistency weight of 0.3. The initial weight matrix is ​​dynamically updated as the data status changes. When data anomalies are detected, the corresponding weight values ​​will be reduced, while the weight values ​​of normal data will be increased accordingly, thus forming the target weight matrix.

[0039] When the target weight matrix is ​​processed by eigendecomposition, the principal component analysis method is used to extract key features. The eigendecomposition process first calculates the data covariance matrix, then calculates the eigenvalues ​​and eigenvectors, and selects the main eigenvectors according to the eigenvalue size. When the eigenvector set is processed by dimensionality reduction, the feature dimensions that need to be retained are determined according to the cumulative contribution rate. Usually, the feature dimensions with a cumulative contribution rate of 95% are retained, and the original high-dimensional data is compressed into a lower-dimensional feature space to obtain a reduced-dimensional feature matrix. The clustering process of the reduced-dimensional feature matrix uses the K-means clustering algorithm. The similarity is calculated based on the Euclidean distance between data points, and similar features are divided into the same group. In the clustering process, the cluster center is continuously adjusted through iterative optimization until the intra-class difference is minimized and the inter-class difference is maximized. After obtaining the preliminary clustering results, the boundary optimization process is performed to calculate the distance between each category, and the boundaries of the categories with closer distances are re-divided to ensure the rationality of the category division and form candidate feature categories.

[0040] In the multimodal fusion processing stage, feature fusion rules are established according to the degree of association between different feature categories. The calculation of the degree of association takes into account indicators such as mutual information and correlation coefficient between features, constructs a feature association matrix, guides the feature fusion process through the association matrix, and finally generates a multimodal feature data stream containing complete information. For example: During a certain flight, data from three sensors, engine speed, fuel flow, and exhaust temperature, were collected within 10 minutes, with a sampling frequency of 1 time per second, for a total of 600 data points. First, the data is divided into two parts according to the flight stage: the climb stage (0-3 minutes) and the cruise stage (3-10 minutes). In the climb stage, the engine speed gradually increases from 50% to 85%, the fuel flow increases from 800kg / h to 2400kg / h, and the exhaust temperature rises from 400℃ to 580℃. Through statistical analysis, it is found that the fuel flow suddenly drops to 500kg / h at the 100th second of the climb stage, which obviously deviates from the normal change trend and is marked as an abnormal point.

[0041] When calculating the weight, since the volatility of the abnormal point is far beyond the normal level, its credibility index is low, and the initial weight is reduced from 1.0 to 0.2. At the same time, since the engine speed and exhaust temperature data are normal, the weight values ​​of these two parameters remain above 0.9. After feature decomposition and dimensionality reduction, the original three parameter data are reduced to two main feature dimensions, which capture 95.8% of the data change information. Cluster analysis divides the data points into two categories: normal climb data group and abnormal data group. After boundary optimization, the abnormal data segment is determined to be the fuel flow data of 95-105 seconds. Through feature fusion, the inherent relationship between engine speed and fuel flow, as well as the law of exhaust temperature change, is comprehensively considered to generate a multimodal feature data stream reflecting the working state of the engine. In this data stream, the abnormal fuel flow data is clearly marked, providing an accurate information basis for subsequent data correction. This example shows the complete processing flow of this method when processing actual flight data, and reflects the effectiveness of the method in anomaly detection and feature extraction.

[0042] In an optional embodiment, the process of executing step S102 may specifically include the following steps:

[0043] (1) Decompose the multimodal feature data stream in the time dimension, segment the data according to the time series sampling interval, obtain the basic time series feature sequence, and perform periodic analysis on the basic time series feature sequence. Calculate the periodic features according to the time series fluctuation law to obtain the time series periodic feature set;

[0044] (2) Perform spatial mapping on the time series period feature set, construct a spatial coordinate system based on the sensor position information, obtain a spatial feature distribution map, and perform regional division on the spatial feature distribution map. Determine the regional boundaries based on the feature density distribution and obtain the feature regional division result.

[0045] (3) Perform hierarchical correlation calculation on the feature region division results, set the correlation threshold according to the feature similarity between regions, obtain the initial correlation matrix, and perform sparse processing on the initial correlation matrix. Select key connections according to the correlation strength to obtain a sparse correlation matrix.

[0046] (4) Perform multi-layer mapping processing on the sparse correlation matrix, build a hierarchical transmission relationship according to the data flow, obtain a hierarchical mapping structure, and perform redundancy elimination processing on the hierarchical mapping structure, remove duplicate connections according to the information duplication, and obtain a simplified mapping relationship;

[0047] (5) The simplified mapping relationship is processed into a network, and the associated network topology is generated according to the node connection relationship to obtain the spatiotemporal feature associated network.

[0048] Optionally, a time dimension decomposition process is performed, and the multimodal feature data stream is orderly segmented according to the sampling interval. The multimodal feature data stream contains multiple types of data such as aircraft attitude, speed, altitude, etc. The sampling interval refers to the acquisition frequency of different types of data, such as attitude data is collected 10 times per second, altitude data is collected 5 times per second, etc. After data segmentation, a basic time series feature sequence is formed, and the sequence is periodically analyzed. The periodic characteristics of the data are identified by Fourier transform, and the amplitude and phase of different frequency components are calculated to obtain a time series periodic feature set that reflects the law of data fluctuation. In the spatial mapping processing stage, the time series periodic feature set is coordinate mapped in combination with the physical layout position of each sensor on the aircraft. The sensor position information includes the three-dimensional coordinates and orientation information of the installation point. The time series data is mapped into the three-dimensional space through coordinate transformation to form a spatial feature distribution map. When the spatial feature distribution map is divided into regions, the density clustering method is used to determine the boundaries of different regions according to the distribution density of feature points. The selection of density threshold is based on the statistical characteristics of the data. The high-density area represents a sensor group with strong feature correlation, and the low-density area represents a sensor combination with weak correlation.

[0049] The feature region division results are processed by hierarchical correlation calculation to calculate the feature similarity between different regions. The similarity calculation considers multiple dimensions, including data change trends, value ranges, fluctuation frequency and other factors. The setting of the correlation threshold is based on historical statistical data. The regions above the threshold are associated to form an initial correlation matrix. The initial correlation matrix is ​​sparsely processed to retain connections with high correlation strength and delete weak connections to obtain a sparse correlation matrix reflecting the main correlation. In the multi-layer mapping processing stage, the sparse correlation matrix constructs a hierarchical structure according to the data flow. The data flow reflects the causal relationship and the direction of influence transmission between parameters. For example, changes in engine speed will affect fuel flow, which in turn affects exhaust temperature. The hierarchical transmission relationship reflects the data transmission path and influence depth, forming a hierarchical mapping structure. The hierarchical mapping structure is de-redundant, and repeated data transmission paths are identified and deleted to ensure that only one optimal path is retained for each association in the network.

[0050] The simplified mapping relationship is processed through network construction, integrating each node and connection relationship into a complete association network. Nodes represent different data features, and connections represent the association relationship between features, ultimately forming a spatiotemporal feature association network that describes the spatiotemporal association characteristics of the data.

[0051] For example, during a flight, multiple sets of sensor data related to the engine are collected. The sampling frequency of the engine speed sensor is 20Hz, the fuel flow sensor is 10Hz, the exhaust temperature sensor is 5Hz, and the vibration sensor is 50Hz. When performing time dimension decomposition, the minimum sampling interval of 0.02 seconds is selected as the benchmark to time align all data. After Fourier analysis, it is found that the engine speed has two main periodic characteristics: low-frequency fluctuations of 0.5Hz and high-frequency vibrations of 5Hz. In the spatial mapping stage, the four sets of sensors are located at different positions of the engine: the speed sensor is located at the bearing box (x=0, y=0, z=0), the fuel flow sensor is located at the inlet of the fuel pipeline (x=0.5, y=0.3, z=0), the exhaust temperature sensor is located at the tail nozzle (x=2.0, y=0, z=0), and the vibration sensor is located on the surface of the casing (x=1.0, y=0.2, z=0.3). According to the distribution density of sensors, the area is divided into three main areas: front, middle, and rear.

[0052] The hierarchical correlation calculation shows that the correlation coefficient between speed and fuel flow is 0.92, the correlation coefficient between speed and exhaust temperature is 0.85, and the correlation coefficient between speed and vibration is 0.78. The correlation threshold is set to 0.8, and the sparse correlation matrix obtained retains the two strong correlation paths of speed-fuel flow and speed-exhaust temperature. Through data flow analysis, it is determined that the response time from speed change to fuel flow change is 0.1 second, and the response time from fuel flow to exhaust temperature is 0.3 second, forming a clear parameter transfer chain. After removing redundant connections, the spatiotemporal feature correlation network finally constructed accurately reflects the correlation and transmission law between engine parameters, providing a reliable feature model for subsequent anomaly detection and data correction.

[0053] In an optional embodiment, the process of executing step S103 may specifically include the following steps:

[0054] (1) Perform data jump detection processing on the spatiotemporal feature association network, calculate the jump point according to the change rate of continuous data points, obtain the rapid warning feature point, and perform noise filtering on the rapid warning feature point, remove the interference point according to the jump amplitude threshold, and obtain the warning data sequence;

[0055] (2) Perform feature pattern matching on the warning data sequence, calculate similarity based on the historical abnormal feature library, obtain pattern matching results, and perform confidence assessment on the pattern matching results. Set confidence levels based on the matching degree to obtain a confidence assessment table.

[0056] (3) Perform semantic rule parsing on the confidence evaluation table, perform constraint checking according to predefined semantic rules, obtain rule verification results, and resolve conflicts on the rule verification results. Adjust conflicting items according to rule priorities to obtain standard verification results.

[0057] (4) The standard verification results are classified into abnormal levels, and the level threshold is set according to the degree of abnormality to obtain an abnormal level table. The abnormal level table is then subjected to correlation analysis. An abnormal chain is constructed based on the correlation between abnormal items to obtain an abnormal correlation map.

[0058] (5) Label the abnormal association graph, generate labeling rules based on the abnormal type and level information, and obtain abnormal labeling data.

[0059] Optionally, data jump detection identifies data anomalies by calculating the rate of change between adjacent data points, and the calculation of the rate of change is based on the ratio of the difference between the data points to the time interval. When the rate of change exceeds a preset threshold, the point is marked as a jump point. For flight parameters, different types of data have different rate of change thresholds, such as the normal rate of change of altitude data does not exceed 50 meters per second, and the normal rate of change of speed data does not exceed 5 meters per second per second. The identified jump points are subjected to noise filtering, and small jumps caused by random interference are removed by setting a jump amplitude threshold, retaining the real abnormal jump points to form an early warning data sequence. The early warning data sequence is processed by feature pattern matching, and the current abnormal pattern is compared with the known pattern in the historical abnormal feature library. The historical abnormal feature library contains feature descriptions of various typical anomalies, such as data performance characteristics in sensor failure, data loss, system interference, etc. The similarity calculation uses multi-dimensional feature comparison, including the matching degree of data waveform shape, amplitude change, duration, etc. The matching results are evaluated for confidence according to the degree of similarity, and the confidence is divided into three levels: high, medium, and low. The similarity greater than 90% is high confidence, 70%-90% is medium confidence, and less than 70% is low confidence. An evaluation table containing anomaly types and confidence levels is generated. The confidence evaluation table enters the semantic rule parsing stage, in which predefined semantic rules are applied to verify the anomalies. Semantic rules are constraints based on flight physics characteristics and system working principles. For example, the altitude should not change significantly in level flight, and the changes in engine parameters should comply with thermodynamic laws. After rule verification, conflicts between different rules are resolved, and the final judgment result is determined through the rule priority mechanism. Rule priority is set based on safety considerations, and rules involving flight safety have higher priority.

[0060] The abnormality level of the standard verification results is based on the severity and impact range of the abnormality. The abnormality level ranges from level 1 to level 4, where level 1 indicates a minor abnormality with no substantial impact on the flight; level 2 indicates a moderate abnormality that requires attention but does not affect normal flight; level 3 indicates a serious abnormality that requires corrective measures; and level 4 indicates a dangerous abnormality that requires immediate processing. The basis for the classification of abnormality levels includes factors such as the degree of data deviation, duration, and impact range. The abnormality level table is subjected to correlation analysis, and by tracing the propagation relationship between abnormalities, an abnormality correlation map reflecting the abnormal propagation path and impact range is constructed. The marking process of the abnormality correlation map systematically marks the type, level, impact range and other information of the abnormality to generate standardized abnormality marking data. The marking rules include information such as the occurrence time of the abnormality, the parameters involved, the abnormality type code, and the severity, which facilitates subsequent data correction and analysis and processing.

[0061] For example, during a flight, the altitude sensor data was abnormal. The original data showed that when the aircraft was cruising at an altitude of 10,000 meters, at a certain moment, the altitude data jumped from 10,000 meters to 12,000 meters within 0.1 seconds, with a change rate of 20,000 meters / second, far exceeding the normal climb rate of 50 meters / second. The data jump detection immediately captured this jump point, and after noise filtering (setting a threshold of 100 meters / second), it was confirmed as an effective warning point. During the feature matching process, the abnormal pattern was compared with the feature library, and it was found that the similarity with the "altitude sensor signal jump failure" pattern reached 95%, and it was assessed as a high-confidence anomaly. Semantic rule verification shows that such a large altitude change during the cruise phase violates the flight physics constraint rules and is inconsistent with the data of other sensors (such as pressure altitude and radar altitude).

[0062] The abnormality level determination classifies it as a level 3 abnormality because: 1) the data deviation exceeds 20% of the normal value; 2) it affects the key parameter of flight altitude; 3) it needs to be corrected in time to ensure navigation accuracy. Correlation analysis found that the abnormality also affected related parameters such as airspeed calculation and vertical speed display, forming an abnormal propagation chain with the altitude sensor as the source point. The final abnormality mark data record is: timestamp 20240315-102030, parameter code AH001 (altitude sensor), abnormality type SC02 (signal jump), abnormality level L3 (level 3), impact range R2 (medium), and related parameters AS001 (airspeed) and VS001 (vertical speed).

[0063] In an optional embodiment, the process of executing step S104 may specifically include the following steps:

[0064] (1) Perform regional boundary division on the abnormal marked data, determine the initial region range according to the spatial adjacent relationship, obtain the abnormal region boundary map, and perform density clustering on the abnormal region boundary map. Merge the adjacent regions according to the distribution density of the abnormal points to obtain the merged region map;

[0065] (2) Recursively segment the merged region graph, set the segmentation threshold according to the degree of anomaly, obtain a multi-level region tree, and perform node screening on the multi-level region tree, remove low-density nodes according to the regional anomaly density, and obtain the core region set;

[0066] (3) Perform diffusion path tracing on the core area set, construct the diffusion direction according to the abnormal transmission relationship between the areas, obtain the diffusion path map, and perform path priority analysis on the diffusion path map, determine the primary and secondary paths according to the diffusion intensity, and obtain a hierarchical path table;

[0067] (4) Perform propagation timing analysis on the hierarchical path table, construct the propagation order according to the time series of anomaly occurrence, obtain the propagation timing chain, and perform causal association analysis on the propagation timing chain, determine the causal terms according to the timing dependency, and obtain the propagation causal network;

[0068] (5) The propagation causal network is reconstructed, and a complete link is generated based on the causal relationship and propagation path to obtain an abnormal propagation link graph.

[0069] Optionally, the spatial adjacency relationship must be clarified before the region boundary division processing of abnormal marked data is performed. The spatial adjacency relationship refers to the physical position association between different sensors on the aircraft. A spatial distance matrix is ​​established based on the three-dimensional coordinate information of the sensor, and sensors that are less than a preset threshold (such as 0.5 meters) apart are identified as adjacent. The initial impact range of the abnormal data is determined by the spatial adjacency relationship, and an abnormal region boundary map is generated. When the abnormal region boundary map is subjected to density clustering processing, a density-based spatial clustering algorithm is used to set a density threshold (such as the number of abnormal points in a unit space is greater than 5), and adjacent regions with similar densities are merged to form a larger continuous abnormal region to obtain a merged region map. The merged region map enters the recursive segmentation processing stage, and a multi-level segmentation threshold is set according to the degree of abnormality. The degree of abnormality is measured by the degree to which the data deviates from the normal value, such as a deviation of 2 times the standard deviation is set as a mild abnormality, 3 times the standard deviation is set as a moderate abnormality, and 4 times the standard deviation is set as a severe abnormality. The recursive segmentation process starts from the top-level region and is segmented downward step by step according to the threshold of the degree of abnormality to form a multi-level region tree with a hierarchical structure. Node screening is performed on the multi-level region tree, the density of abnormal points in each node area is calculated, and nodes with abnormal point density lower than the threshold (such as density less than 3 points / cubic meter) are removed, and the core region set reflecting the main abnormal characteristics is retained.

[0070] The core area set is processed by diffusion path tracing to analyze the transmission relationship between anomalies in different areas. The transmission relationship is determined based on the physical connection and logical dependency of the data, such as the transmission of engine parameter anomalies to adjacent temperature sensors, which then affect the pressure sensor. The diffusion direction is constructed based on the transmission relationship, and the propagation path of the anomaly from the source point to each affected area is recorded to obtain a diffusion path map. The diffusion path map is prioritized and the diffusion intensity on each propagation path is calculated. The intensity value is determined based on the degree of anomaly and the propagation distance. Based on this, the path is divided into primary path and secondary path to generate a hierarchical path table. In the propagation timing analysis processing stage of the hierarchical path table, the specific time points when the anomaly appears in each area are recorded, and the propagation timing chain is constructed in chronological order. The propagation timing chain reflects the complete time process from the initial appearance to the diffusion of the anomaly. The causal relationship is determined based on the physical relationship and timing between the parameters, the source of the anomaly and the affected parameter chain are identified, and the propagation causal network is constructed.

[0071] The link reconstruction process of the propagation causal network is to integrate all the identified causal relationships and propagation paths into a complete abnormal propagation link. Link reconstruction takes into account information in multiple dimensions such as temporal order, spatial location, and parameter dependency, and generates a complete abnormal propagation link diagram, which clearly shows the propagation process and impact range of the abnormality.

[0072] For example, during the flight of an aircraft, the temperature sensor of the right engine detected an abnormal high temperature signal. First, the regional boundary was divided, and it was found that 8 related sensors were arranged within a radius of 0.8 meters in the engine compartment, including temperature, pressure, vibration and other types. Density clustering analysis showed that the density of abnormal points within 0.5 meters around the temperature sensor reached 8 / cubic meter, exceeding the set threshold of 5 / cubic meter, and this area was marked as the initial abnormal area. During the recursive segmentation process, it was found that the temperature abnormal value reached 2.5 times the normal value (a moderate abnormality). Based on this threshold, the abnormal area was divided into three layers: core layer (temperature sensor location, abnormal value > 2.5 times), middle layer (within 0.3 meters, abnormal value 1.8-2.5 times), and peripheral layer (within 0.3-0.8 meters, abnormal value 1.2-1.8 times). Node screening removed the areas in the peripheral layer with an abnormal density less than 3 / cubic meter, and retained the core layer and middle layer as the core area set. Diffusion path tracing found that the anomaly started from the temperature sensor and affected the adjacent pressure sensor (temperature increase caused pressure change), fuel flow sensor (temperature increase affected fuel vaporization) and vibration sensor (temperature change caused mechanical property change) in 10 seconds. The main path was the temperature-pressure-fuel flow sensor link (anomaly intensity 0.8), and the secondary path was the temperature-vibration sensor link (anomaly intensity 0.4).

[0073] The timing analysis shows that the order of abnormal propagation is: 0 seconds of temperature abnormality (850°C, 300°C above the standard), 2 seconds of pressure abnormality (15% decrease), 5 seconds of fuel flow abnormality (10% decrease), 8 seconds of vibration abnormality (20% increase). The causal analysis confirmed that the temperature abnormality was the source. The temperature change affected the pressure through the thermal effect. The pressure change caused the fuel flow abnormality, which eventually caused the mechanical vibration abnormality. The integrated abnormal propagation link diagram fully records the whole process of this abnormality from occurrence to spread.

[0074] In an optional embodiment, the process of executing step S105 may specifically include the following steps:

[0075] (1) Extract key nodes from the anomaly propagation link graph, calculate the importance value based on the node connectivity, obtain the key node sequence, and perform a correction priority sorting process on the key node sequence. Set the priority level according to the anomaly impact range to obtain the priority node table.

[0076] (2) Perform data missing analysis on the priority node table, mark missing segments according to data integrity requirements, obtain a missing data list, and perform reconstruction parameter setting on the missing data list. Divide the reconstruction rules according to the data type to obtain a parameter reconstruction table;

[0077] (3) Perform data completion processing on the parameter reconstruction table, generate completion data according to the reconstruction rules, obtain a data reconstruction sequence, perform physical constraint verification processing on the data reconstruction sequence, correct the completion value according to the flight dynamics rules, and obtain a corrected data set;

[0078] (4) Perform impact range analysis on the calibration data set, calculate the impact area based on the data correlation, obtain the impact assessment map, and perform risk assessment on the impact assessment map. Set the risk level according to the impact degree to obtain a risk level table;

[0079] (5) Plan the correction steps for the risk level table, generate operation steps based on the risk level and correction rules, and obtain the data correction execution sequence.

[0080] Optionally, a key node refers to a data point that has an important influence in the process of anomaly propagation, and the node connectivity refers to the number of other nodes directly connected to the node. The calculation of the importance value comprehensively considers the node connectivity, anomaly degree and parameter type. The higher the connectivity, the greater the anomaly degree and the more important the parameter, the higher its importance value. By setting the importance threshold, the key nodes are screened out to form a key node sequence. When the key node sequence is prioritized, the impact range of the anomaly is considered, including the number of directly affected parameters, the system functions indirectly affected, the degree of threat to flight safety, etc., and the nodes are divided into different priority levels based on this to generate a priority node table. The priority node table enters the data missing analysis stage, and the data status of each node is checked according to the data integrity requirements. The data integrity requirements include sampling frequency requirements, numerical accuracy requirements, time continuity requirements, etc. When the data does not meet the integrity requirements, it is marked as a missing fragment, and the missing time period, data type, missing reason and other information are recorded to form a missing data list. The missing data list is reconstructed with parameter settings, and different reconstruction rules are formulated for different types of data. The reconstruction rules include interpolation method selection, boundary condition setting, physical constraint conditions, etc., and a parameter reconstruction table is generated according to these rules.

[0081] The parameter reconstruction table guides the data completion process and completes the missing data according to the set reconstruction rules. The completion methods include linear interpolation, spline interpolation, physical model extrapolation, etc. Select the appropriate method to generate the completed data and form a data reconstruction sequence. Perform physical constraint inspection on the data reconstruction sequence and apply flight dynamics rules to verify the rationality of the completed data. The dynamics rules include physical laws such as velocity-acceleration relationship, attitude angle constraint, and energy conservation to ensure that the completed data conforms to physical laws and generate a correction data set. The impact range analysis of the correction data set determines the impact range of the correction operation by calculating the data correlation. The data correlation reflects the mutual dependence between parameters. The correlation calculation considers the physical correlation, statistical correlation and time series dependence between parameters. The boundary of the impact area is determined according to the strength of the correlation, and an impact assessment map is generated. Risk assessment is performed on the impact assessment map. The risk level is assessed according to the severity, duration and diffusion trend of the impact. The risk is divided into three levels: low, medium and high to form a risk level table.

[0082] In the correction step planning phase of the risk level table, detailed correction operation steps are formulated according to the risk level and the complexity of the correction rules. Correction rules include data replacement rules, smooth transition rules, consistency maintenance rules, etc. These rules are organized into an orderly operation sequence to form a data correction execution sequence.

[0083] For example, during a certain flight, the attitude measurement system had an abnormality. The abnormal propagation link diagram shows that the roll angle sensor is the source of the abnormality, with a connectivity of 5 (connected to the yaw angle, angle of attack, sideslip angle, and two acceleration sensors). The importance value calculation shows that the importance of the roll angle node is 0.85 (connectivity contribution 0.35, abnormality contribution 0.3, parameter importance contribution 0.2), which exceeds the threshold of 0.6 and is listed as the highest priority correction node. Data missing analysis found that the roll angle data was intermittently lost within 10 seconds (t=100s to t=110s). The normal sampling frequency should be 50Hz, but only 10 valid sample points were actually obtained, forming a missing data list. The reconstruction parameter setting is based on the roll angle characteristics, selects the cubic spline interpolation method, and sets physical constraints: the roll angle change rate does not exceed 30 degrees / second, and the roll angle range is within ±60 degrees.

[0084] The data completion process is based on the above rules, using valid data points before and after the moment for interpolation and reconstruction. The original 10 valid sample points are: 100s (15 degrees), 101s (20 degrees), 103s (missing), 104s (35 degrees), 105s (missing), 106s (45 degrees), 107s (missing), 108s (30 degrees), 109s (25 degrees), 110s (20 degrees). Through cubic spline interpolation, a complete 50Hz data sequence is reconstructed, and physical constraint inspection is carried out to ensure that the reconstructed data meets the law of roll angle changes. Impact range analysis shows that roll angle correction will directly affect yaw angle calculation (correlation 0.8) and sideslip angle calculation (correlation 0.7). Risk assessment believes that this impact level belongs to medium risk (level 2), because although it affects multiple attitude parameters, the corrected data is still within the controllable range.

[0085] The final correction execution sequence includes: 1. Roll angle data reconstruction (t=100-110s); 2. Yaw angle data synchronization correction (t=100-112s, considering the 2-second delay effect); 3. Sideslip angle data synchronization correction (t=100-112s); 4. Attitude parameter consistency check. This execution sequence ensures that the attitude data correction process is carried out in an orderly manner and maintains the physical relationship between parameters.

[0086] In an optional embodiment, the process of executing step S106 may specifically include the following steps:

[0087] (1) The data correction execution sequence is divided into segment points, and the checkpoint position is set according to the complexity of data correction to obtain a segment check sequence. The segment check sequence is then processed into a state snapshot, and the data state information is recorded according to the checkpoint to obtain a state snapshot table.

[0088] (2) Decompose the state snapshot table into correction steps, divide the sub-steps according to the dependency relationship of the correction operation, obtain the correction sub-sequence, and perform sequential verification on the correction sub-sequence. Adjust the execution order according to the constraint relationship between the steps to obtain the verification sequence table;

[0089] (3) performing incremental correction processing on the verification sequence table, performing data correction according to the sub-step sequence to obtain incremental correction data, and performing consistency verification processing on the incremental correction data, verifying the correction results according to the physical constraint rules, and obtaining a verification result set;

[0090] (4) Perform rollback point determination processing on the inspection result set, determine the rollback position according to the inspection status, obtain a rollback point list, and perform state recovery processing on the rollback point list, restore the data state according to the rollback rules, and obtain a recovery data set;

[0091] (5) Perform integrity assessment on the restored data set, evaluate the correction effect based on the integrity index, and obtain the corrected flight data.

[0092] Optionally, the complexity of the correction is determined by multiple factors, including the number of correction parameters, the correlation between parameters, the complexity of the correction algorithm, etc. Checkpoints need to be set at locations with high complexity. The density of checkpoints is proportional to the complexity. Key checkpoints are usually set at locations where parameters change drastically and where multiple parameters are coupled. After the segmented check sequence is formed, the complete state information of the data is recorded at each checkpoint, including parameter values, correction marks, and associated parameter states. This information constitutes a state snapshot table, which is used to track data changes during the correction process. The state snapshot table is decomposed by the correction steps to decompose complex correction operations into multiple simple sub-steps. The decomposition process is based on the dependencies between correction operations, which include data dependencies (the correction of a parameter depends on the values ​​of other parameters) and timing dependencies (correction operations need to be performed in a specific order). Each sub-step has a clear input and output definition to form a correction sub-sequence. The correction sub-sequence is sequentially verified to ensure that the execution order of the steps meets various constraints, such as causal constraints, resource usage constraints, etc., and finally a verified sequence table is generated.

[0093] The verification sequence table enters the incremental correction processing stage, and the correction operation is performed step by step in the order after verification. Incremental correction means that only a small part of the data is corrected each time, and the correction result is immediately checked for consistency. The consistency check is based on physical constraint rules, including numerical range constraints, change rate constraints, parameter relationship constraints, etc. The correction results that pass the test are saved to the test result set, and the results that fail the test will trigger the rollback mechanism. The rollback point determination process of the test result set determines the location that needs to be rolled back based on the test status. The test status includes the data consistency level, the degree of satisfaction of physical constraints, the degree of influence of associated parameters, etc. When the test status shows that the correction result does not meet the requirements, it is necessary to determine the specific location of the rollback, and usually the nearest stable state point is selected as the rollback target. The rollback point list records all potential rollback locations and their state information, and performs state recovery according to the rollback rules to restore the data to the specified historical state to form a recovery data set.

[0094] The restored data set undergoes integrity assessment, which includes multiple dimensions such as data continuity, consistency, and accuracy. The integrity index quantifies the correction effect and ultimately generates corrected flight data that meets the requirements.

[0095] For example, during a flight, the pressure altitude data is abnormal and needs to be corrected. The data correction execution sequence is first divided into multiple check segments. The complexity analysis of altitude data correction shows that the complexity is higher in the intervals where the altitude changes rapidly (such as the climb and descent segments), so the density of checkpoints is increased in these intervals. The specific checkpoint locations are: after takeoff (t=60s), mid-climb (t=180s), start of cruise (t=300s), start of descent (t=1200s), and before landing (t=1500s). The status snapshot is recorded at each checkpoint, including: at t=60s, the pressure altitude is 300 meters, the ascent speed is 5 meters / second, the corresponding atmospheric pressure is 980 hPa, and the temperature is 15 degrees Celsius; at t=180s, the pressure altitude is 3000 meters, the ascent speed is 10 meters / second, the atmospheric pressure is 700 hPa, and the temperature is 5 degrees Celsius, and so on. The correction steps are broken down into: first correct the pressure altitude data, then update the atmospheric pressure value, and finally calibrate the temperature correction value. Verify the order in which these steps are executed, ensuring that the correction of the basic parameters (pressure altitude) is completed first, and then the updated derived parameters (barometric pressure, temperature) are performed.

[0096] During the incremental correction process, the data in the altitude range of 100 meters is corrected each time. For example, at t=180s, the original pressure altitude is 3000 meters, but it differs from the value displayed by the radio altimeter by 200 meters. The correction is performed by segmented interpolation, with 20 meters corrected each time, and completed in 10 times. A consistency check is performed immediately after each correction to verify whether the corrected altitude data meets the following requirements: 1) the altitude change rate does not exceed 15 meters / second; 2) the difference with the radio altitude is within the range of ±50 meters; 3) the corresponding atmospheric pressure change conforms to the standard atmosphere model. At t=250s, the correction result shows a conflict with other constraints, triggering the rollback mechanism. The rollback point determination determines that it is necessary to roll back to the state of t=240s (the nearest stable point). The state recovery process cancels all correction results from t=240s to t=250s, restores them to the original state, and then re-executes the correction operation using a new correction strategy.

[0097] The final integrity assessment showed that the continuity of the corrected pressure altitude data reached 99.9%, and the average consistency error with the radio altitude was less than 30 meters, meeting the accuracy requirements of flight data recording. The entire correction process maintained the physical rationality of the data while ensuring the coordination and consistency of related parameters.

[0098] The above describes the integrity check and correction method for flight data in the embodiment of the present application. The following describes the integrity check and correction system for flight data in the embodiment of the present application. Figure 2 In one embodiment of the present application, a system for integrity checking and correction of flight data includes:

[0099] The acquisition module is used to perform distributed acquisition and processing of flight data, filter data through a dynamic weight self-adjustment mechanism, and obtain a multi-modal feature data stream;

[0100] The extraction module is used to perform spatiotemporal feature extraction processing on the multimodal feature data stream, construct an association matrix based on the multi-level data mapping relationship, and obtain a spatiotemporal feature association network;

[0101] The verification module is used to perform hierarchical verification processing on the spatiotemporal feature association network, perform data verification according to the three-level architecture of rapid warning, pattern recognition, and semantic verification, and obtain abnormal labeled data;

[0102] The positioning module is used to implement regional recursive positioning processing on the abnormal marked data, combining multi-level threshold filtering and regional diffusion tracking to obtain the abnormal propagation link map;

[0103] A processing module is used to design and process a correction scheme for the abnormal propagation link diagram, and obtain a data correction execution sequence based on data reconstruction and impact assessment optimization scheme;

[0104] The correction module is used to implement progressive processing on the data correction execution sequence, use segmented status checking and dynamic rollback mechanism to correct the data, and obtain corrected flight data.

[0105] Through the coordinated cooperation of the above-mentioned components, the risk of single-point collection failure is avoided and the reliability and robustness of data collection are improved by distributing the collection and processing of flight data and introducing a dynamic weight self-adjustment mechanism. The spatiotemporal feature extraction of multimodal feature data streams and the construction of multi-level data mapping relationships realize the comprehensive capture and in-depth analysis of data features, providing a reliable feature basis for anomaly detection. The three-level architecture of rapid warning, pattern recognition and semantic verification is used for hierarchical inspection to build a multi-level anomaly detection defense line, which significantly improves the accuracy and reliability of anomaly detection. The spatial distribution characteristics of abnormal data are accurately identified by combining regional recursive positioning and multi-level threshold filtering, and the propagation law and impact range of anomalies are accurately grasped by combining the regional diffusion tracking mechanism. In the design of the correction plan, the dual guarantee of data reconstruction and impact assessment is used to ensure the scientificity and feasibility of the correction plan, and effectively reduce the negative impact of the correction process on related data. In the process of correction implementation, a progressive processing strategy is adopted to timely discover problems through segmented status inspection, and errors are corrected in time by relying on a dynamic rollback mechanism, which greatly improves the reliability and stability of data correction. The overall solution forms a complete data processing closed loop, with each processing link closely connected and mutually supporting, which not only ensures the integrity and accuracy of flight data, but also provides a more reliable data basis for flight safety analysis. Especially in terms of anomaly detection and data correction, through multi-level verification and progressive correction strategies, it not only ensures the accuracy of processing, but also ensures the controllability of the correction process.

[0106] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for integrity checking and correction of flight data.

[0107] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for integrity checking and correction of flight data, characterized in that: The integrity check and correction method for flight data includes: Perform distributed collection and processing of flight data, filter data through a dynamic weight self-adjustment mechanism, and obtain multi-modal feature data streams; Perform spatiotemporal feature extraction on the multimodal feature data stream, build a correlation matrix based on the multi-level data mapping relationship, and obtain a spatiotemporal feature correlation network; Perform hierarchical inspection on the spatiotemporal feature association network, perform data verification according to the three-level architecture of rapid warning, pattern recognition, and semantic verification, and obtain abnormal labeled data; Implement regional recursive positioning processing on abnormal labeled data, combine multi-level threshold filtering and regional diffusion tracking to obtain an abnormal propagation link map; The correction scheme design is processed for the abnormal propagation link graph, and the data correction execution sequence is obtained based on the data reconstruction and impact assessment optimization scheme; The data correction execution sequence is processed progressively, and the data is corrected using segmented status checking and dynamic rollback mechanism to obtain corrected flight data.

2. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The distributed collection and processing of flight data, screening data through a dynamic weight self-adjustment mechanism, and obtaining a multi-modal feature data stream include: Performing time segmentation processing on the flight data, performing data segmentation according to flight phase characteristics to obtain an initial time series data set, and performing outlier marking processing on the initial time series data set, setting a threshold boundary according to data distribution characteristics, and obtaining marked time series data; Performing weight calculation processing on the marked time series data, calculating weight coefficients according to data credibility indicators to obtain an initial weight matrix, and dynamically updating the initial weight matrix, adjusting weight values ​​according to data fluctuation status to obtain a target weight matrix; Performing feature decomposition processing on the target weight matrix, extracting key features according to principal component analysis of the data to obtain a feature vector set, and performing dimensionality reduction processing on the feature vector set, selecting feature dimensions according to the information retention rate, and obtaining a reduced-dimensional feature matrix; Performing clustering processing on the dimension-reduced feature matrix, dividing feature groups according to data similarity to obtain feature clustering results, and performing boundary optimization processing on the feature clustering results, adjusting cluster boundaries according to inter-class distances to obtain candidate feature categories; The candidate feature categories are subjected to multimodal fusion processing, and fusion rules are constructed according to the feature category associations to obtain a multimodal feature data stream.

3. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The spatiotemporal feature extraction process is performed on the multimodal feature data stream, and a correlation matrix is ​​constructed according to the multi-level data mapping relationship to obtain a spatiotemporal feature correlation network, including: Performing time dimension decomposition processing on the multimodal feature data stream, performing data segmentation according to the time series sampling interval to obtain a basic time series feature sequence, and performing periodic analysis processing on the basic time series feature sequence, calculating the periodic features according to the time series fluctuation law, and obtaining a time series periodic feature set; Performing spatial mapping processing on the time series period feature set, constructing a spatial coordinate system according to the sensor position information, obtaining a spatial feature distribution map, and performing region division processing on the spatial feature distribution map, determining region boundaries according to feature density distribution, and obtaining a feature region division result; Performing hierarchical correlation calculation processing on the feature region division results, setting correlation thresholds according to the feature similarity between regions to obtain an initial correlation matrix, and performing sparse processing on the initial correlation matrix, screening key connections according to correlation strength, and obtaining a sparse correlation matrix; Performing multi-layer mapping processing on the sparse association matrix, constructing a hierarchical transmission relationship according to the data flow direction to obtain a hierarchical mapping structure, and performing redundancy elimination processing on the hierarchical mapping structure to remove repeated connections according to information duplication to obtain a simplified mapping relationship; The simplified mapping relationship is subjected to network construction processing, and an associated network topology is generated according to the node connection relationship to obtain a spatiotemporal feature associated network.

4. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The hierarchical inspection process is performed on the spatiotemporal feature association network, and data verification is performed according to the three-level architecture of rapid early warning, pattern recognition, and semantic verification to obtain abnormal labeled data, including: Performing data jump detection processing on the spatiotemporal feature association network, calculating the jump point according to the change rate of continuous data points, obtaining rapid warning feature points, and performing noise filtering processing on the rapid warning feature points, removing interference points according to the jump amplitude threshold, and obtaining a warning data sequence; Performing feature pattern matching processing on the warning data sequence, performing similarity calculation based on the historical abnormal feature library to obtain a pattern matching result, and performing confidence evaluation processing on the pattern matching result, setting confidence levels according to the matching degree, and obtaining a confidence evaluation table; Performing semantic rule parsing processing on the confidence evaluation table, performing constraint checking according to predefined semantic rules to obtain rule verification results, performing conflict resolution processing on the rule verification results, adjusting conflicting items according to rule priorities, and obtaining standard verification results; The standard verification results are processed by abnormal level classification, and level thresholds are set according to the degree of abnormality to obtain an abnormal level table, and the abnormal level table is processed by association analysis, and an abnormal chain is constructed according to the association relationship between abnormal items to obtain an abnormal association map; The abnormal association map is labeled, and a labeling rule is generated according to the abnormal type and level information to obtain abnormal labeling data.

5. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The abnormality marked data is subjected to regional recursive positioning processing, combined with multi-level threshold filtering and regional diffusion tracking, to obtain an abnormality propagation link diagram, including: Performing a regional boundary division process on the abnormal marked data, determining the initial region range according to the spatial adjacent relationship, obtaining an abnormal region boundary map, and performing a density clustering process on the abnormal region boundary map, merging adjacent regions according to the abnormal point distribution density, and obtaining a merged region map; Recursively segmenting the merged region graph, setting a segmentation threshold according to the degree of abnormality, obtaining a multi-level region tree, and performing node screening on the multi-level region tree, removing low-density nodes according to regional abnormality density, and obtaining a core region set; Performing diffusion path tracing processing on the core area set, constructing diffusion directions according to the abnormal transmission relationship between areas, obtaining a diffusion path map, and performing path priority analysis processing on the diffusion path map, determining primary and secondary paths according to diffusion intensity, and obtaining a hierarchical path table; Performing a propagation timing analysis on the hierarchical path table, constructing a propagation sequence according to the abnormal occurrence time series, obtaining a propagation timing chain, and performing a causal association analysis on the propagation timing chain, determining causal terms according to the timing dependency, and obtaining a propagation causal network; The propagation causal network is subjected to link reconstruction processing, and a complete link is generated according to the causal relationship and the propagation path to obtain an abnormal propagation link graph.

6. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The correction scheme design process for the abnormal propagation link diagram is performed based on data reconstruction and impact assessment optimization scheme to obtain a data correction execution sequence, including: Extract key nodes from the abnormal propagation link graph, calculate importance values ​​based on node connectivity, obtain a key node sequence, perform a modified priority sorting process on the key node sequence, set a priority level based on the abnormal impact range, and obtain a priority node table; Performing data missing analysis processing on the priority node table, marking missing segments according to data integrity requirements, obtaining a missing data list, and performing reconstruction parameter setting processing on the missing data list, dividing reconstruction rules according to data types, and obtaining a parameter reconstruction table; Performing data completion processing on the parameter reconstruction table, generating completion data according to a reconstruction rule to obtain a data reconstruction sequence, performing physical constraint inspection processing on the data reconstruction sequence, correcting the completion value according to a flight dynamics rule to obtain a correction data set; Performing an impact range analysis on the correction data set, calculating the impact area according to the data correlation, obtaining an impact assessment map, and performing a risk assessment on the impact assessment map, setting a risk level according to the impact degree, and obtaining a risk level table; The risk level table is processed by correction step planning, and operation steps are generated according to the risk level and correction rules to obtain a data correction execution sequence.

7. The method for integrity checking and correction of flight data according to claim 1, characterized in that: The step of implementing progressive processing on the data correction execution sequence, using segmented status checking and dynamic rollback mechanism to correct the data, and obtaining corrected flight data, includes: The data correction execution sequence is processed by segmentation points, and the checkpoint positions are set according to the complexity of the data correction to obtain a segmented check sequence, and the segmented check sequence is processed by state snapshot, and data state information is recorded according to the checkpoints to obtain a state snapshot table; The state snapshot table is subjected to correction step disassembly processing, and sub-steps are divided according to the dependency relationship of the correction operation to obtain a correction subsequence, and the correction subsequence is subjected to sequential verification processing, and the execution order is adjusted according to the constraint relationship between the steps to obtain a verification sequence table; Performing incremental correction processing on the verification sequence table, performing data correction according to the sub-step sequence to obtain incremental correction data, and performing consistency verification processing on the incremental correction data, verifying the correction result according to the physical constraint rule, and obtaining a verification result set; Performing rollback point determination processing on the inspection result set, determining the rollback position according to the inspection status, obtaining a rollback point list, and performing state recovery processing on the rollback point list, restoring the data state according to the rollback rule, and obtaining a recovery data set; An integrity assessment process is performed on the restored data set, and a correction effect is evaluated according to an integrity index to obtain corrected flight data.

8. A system for checking and correcting integrity of flight data, used to implement the method for checking and correcting integrity of flight data as claimed in any one of claims 1 to 7, characterized in that: The integrity check and correction system for flight data includes: The acquisition module is used to perform distributed acquisition and processing of flight data, filter data through a dynamic weight self-adjustment mechanism, and obtain a multi-modal feature data stream; The extraction module is used to perform spatiotemporal feature extraction processing on the multimodal feature data stream, construct an association matrix based on the multi-level data mapping relationship, and obtain a spatiotemporal feature association network; The verification module is used to perform hierarchical verification processing on the spatiotemporal feature association network, perform data verification according to the three-level architecture of rapid warning, pattern recognition, and semantic verification, and obtain abnormal labeled data; The positioning module is used to implement regional recursive positioning processing on the abnormal marked data, combining multi-level threshold filtering and regional diffusion tracking to obtain the abnormal propagation link map; A processing module is used to design and process a correction scheme for the abnormal propagation link diagram, and obtain a data correction execution sequence based on data reconstruction and impact assessment optimization scheme; The correction module is used to implement progressive processing on the data correction execution sequence, use segmented status checking and dynamic rollback mechanism to correct the data, and obtain corrected flight data.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for integrity checking and correction of flight data as described in any one of claims 1 to 7 is implemented.

Images