An abnormal flight intelligent identification method based on low-altitude monitoring data

Abstract

The application discloses an abnormal flight intelligent identification method based on low-altitude monitoring data, and comprises the following steps: S1, collecting multi-source low-altitude monitoring data; S2, dividing into multiple sub-trajectory segments according to a flight event timestamp and a preset time window, performing interpolation point supplementing and smoothing processing; S3, calculating a physical quantity perception distance of any two sub-trajectory segments and standardizing to form a physical quantity perception distance matrix; S4, performing abnormal detection by using an improved TRAOD algorithm, and outputting a local outlier index and an abnormal tendency signal; S5, calculating a penalty term based on a flight physical envelope and airspace management rules; S6, constructing an abnormal comprehensive score function, combining the local outlier index and the penalty term by weighting, and introducing the abnormal tendency signal dynamic adjustment, and determining abnormal flight behavior when the comprehensive abnormal score exceeds a threshold value; and S7, generating an abnormal flight identification report. The application realizes high-precision identification of abnormal flight behavior in a low-altitude monitoring environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent anomaly detection and flight behavior recognition technology, and in particular to an intelligent recognition method for abnormal flight based on low-altitude monitoring data. Background Technology

[0002] With the rapid growth of the low-altitude economy, drone regulation, and urban airspace safety demands, intelligent monitoring and anomaly identification technologies for low-altitude aircraft operation have become a research hotspot. Existing low-altitude monitoring systems typically rely on single-source observation equipment, such as radar monitoring systems or ADS-B receiving systems, to identify and record the trajectories of flying targets. However, in complex urban environments and densely populated low-altitude airspace, a single data source cannot comprehensively cover the dynamic characteristics of targets and is easily affected by factors such as terrain obstruction, electromagnetic interference, and weather changes, leading to problems such as discontinuity, offset, and excessive noise in the monitoring data. Furthermore, the sampling frequency, time reference, and coordinate system differences between multi-source monitoring data are significant, making it difficult for traditional data alignment and fusion methods to achieve high-precision synchronization, resulting in accumulated trajectory stitching errors and unstable anomaly detection results.

[0003] In low-altitude flight anomaly identification, existing technologies mainly rely on statistical features or machine learning models for pattern classification. However, most methods depend on historical label samples or global thresholds for judgment, making it difficult to adapt to the changing distribution of anomalies in dynamic environments. Especially when aircraft trajectory data has significant nonlinear and non-stationary characteristics, traditional Euclidean distance or dynamic time warping (DTW) similarity measures cannot fully reflect the comprehensive differences in multidimensional physical quantities such as speed, altitude, and heading angle of the flight trajectory, resulting in insufficient anomaly identification accuracy. Some studies have attempted to introduce clustering and density detection algorithms, such as LOF and DBSCAN, but these algorithms have high computational complexity in high-dimensional trajectory spaces, are sensitive to neighborhood parameters, and lack the ability to process real-time streaming data, failing to meet the real-time requirements of low-altitude monitoring scenarios.

[0004] Furthermore, existing methods typically ignore the correlation between flight behavior and airspace rules and physical constraints. In real-world operating environments, aircraft may experience short-term attitude fluctuations while adhering to flight path rules, or they may overstep boundaries in no-fly zones or altitude-restricted areas. However, traditional algorithms cannot distinguish between physical anomalies and rule-breaking anomalies, leading to a high false alarm rate. Additionally, anomaly detection results are often output as outliers or labels, lacking structured management and visualization of abnormal events, which hinders regulatory authorities from conducting source tracing and statistical analysis.

[0005] Therefore, how to provide an intelligent identification method for abnormal flights based on low-altitude monitoring data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent identification method for abnormal flights based on low-altitude monitoring data. This invention combines multi-source low-altitude monitoring data fusion with an improved TRAOD algorithm to establish a unified time benchmark for data synchronization and standardized processing. It constructs a trajectory similarity model based on physical quantity sensing distance, introduces spatial indexing and event anchor point segmentation mechanisms to achieve real-time anomaly detection of trajectory flow, and calculates penalty terms by combining flight physical envelope and airspace rules and weighted to generate a comprehensive anomaly score. Ultimately, it achieves accurate identification of low-altitude flight behavior and outputs a visual report.

[0007] An intelligent identification method for abnormal flight based on low-altitude monitoring data according to an embodiment of the present invention includes the following steps: S1. Collect multi-source low-altitude monitoring data, and perform time base synchronization and data format unification processing on the multi-source low-altitude monitoring data to form a synchronized monitoring dataset. S2. Based on the flight event timestamp and preset time window, the synchronized monitoring dataset is divided into multiple sub-trajectory segments, and interpolation and smoothing are performed on each sub-trajectory segment. S3. Calculate the physical quantity sensing distance for any two sub-trajectory segments and perform standardization to form a physical quantity sensing distance matrix; S4. An improved TRAOD algorithm based on the physical quantity sensing distance matrix is ​​used to detect anomalies in each sub-trajectory segment and output local outlier index and anomaly tendency signal. S5. Calculate penalty terms for sub-trajectory segments based on flight physical envelope and airspace management rules; S6. Construct an abnormal comprehensive scoring function. Based on the abnormal tendency signal, combine the local outlier index and the penalty term according to a preset weight to obtain a comprehensive abnormal score. Calculate the quantile threshold. When the comprehensive abnormal score exceeds the corresponding quantile threshold, determine that the sub-trajectory segment is an abnormal flight behavior. S7. Generate an abnormal flight identification report and store the report in the monitoring database.

[0008] Optionally, step S1 includes: S11. Deploy multi-source monitoring equipment in the low-altitude monitoring area. The multi-source monitoring equipment includes radar observation equipment, ADS-B receiver, photoelectric sensor and wireless direction finding device. Each device outputs the original observation data of the flying target according to the preset sampling frequency. S12. Collect the raw observation data and record the equipment identifier, flight event timestamp, monitoring node coordinates, target plane coordinates, altitude, heading angle, speed, signal strength, and data quality markers; S13. Establish a unified time reference, which is generated based on BeiDou or GPS timing signals and synchronizes each monitoring node through a network clock protocol, recording the synchronization status and clock offset. S14. Perform time base synchronization and use sampling time interpolation and alignment strategies to eliminate time drift; S15. Perform format unification processing, which includes field name standardization, measurement unit conversion, missing value placeholder filling and sensor type labeling. S16. The multi-source observation data that has been synchronized in time and formatted is aggregated into a synchronized monitoring dataset.

[0009] Optionally, step S2 includes: S21. The synchronized monitoring dataset is divided into multiple sub-trajectory segments based on the flight event timestamp and a preset time window. The preset time window is determined according to the target flight speed and sampling frequency. S22. Perform data sorting on each sub-trajectory segment after division, and arrange them in ascending order according to the flight event timestamp to form a time-continuous sequence of trajectory points; S23. Perform interpolation to fill missing time periods in the trajectory point sequence. The interpolation is based on the spatial coordinates, velocity and heading angle of adjacent valid trajectory points to calculate the position of the missing points. S24. Perform smoothing on the trajectory point sequence after interpolation and use a weighted average method within a sliding time window to suppress noise deviation. S25. Assign a unique identifier to each trajectory segment, and record the trajectory start and end times, number of samples, and monitoring source composition information, and store them as a standardized trajectory dataset.

[0010] Optionally, step S3 includes: S31. For any two sub-trajectory segments in the standardized trajectory dataset, extract a set of trajectory points for distance calculation based on a uniform sampling interval. The set of trajectory points includes planar coordinates, altitude, velocity, and heading angle information. S32. Calculate the spatial path similarity between two sub-trajectory segments. The spatial path similarity is determined based on the sequence of changes in the plane coordinates of the trajectory points and the continuous trend of changes in the heading angle. S33. Calculate the differences between the two sub-track segments in terms of speed, altitude and heading angle, and obtain the average speed difference, average altitude difference and average heading angle difference respectively. S34. Based on the spatial path similarity, average speed difference, average altitude difference and average heading angle difference, set weighting coefficients and perform weighted combination to obtain the physical quantity sensing distance; S35. Perform standardization processing on the physical quantity sensing distance results to eliminate the dimensional differences of different physical quantities and form a comparable physical quantity sensing distance matrix.

[0011] Optionally, step S4 includes: S41. Anomaly detection is performed on the physical quantity sensing distance matrix using an improved TRAOD algorithm. The improved TRAOD algorithm includes four stages: spatial index construction, local density calculation, outlier calculation, and incremental update. S42. In the spatial index construction stage, a spatial index structure is established based on the distance relationship between trajectory segments in the physical quantity sensing distance matrix. The spatial index structure is constructed by combining grid partitioning and neighborhood radius search. S43. In the local density calculation stage, the local density value of the target trajectory segment is calculated based on the distance between the target trajectory segment and the corresponding trajectory segment in the physical quantity sensing distance matrix. S44. In the outlier calculation stage, the neighborhood density value is calculated. The neighborhood density value is the average value of the local density of each trajectory segment in the neighborhood set of the target trajectory segment. The local outlier index is determined based on the difference between the local density and the neighborhood density of the target trajectory segment. S45. In the incremental update phase, the continuous trajectory data is processed in batches based on the event anchor point segmentation structure. The event anchor point segmentation structure divides the trajectory into segments according to the event time anchor points set by the rate of change of heading angle, the rate of change of velocity, or the rate of change of altitude in the trajectory. Neighborhood incremental update and outlier correction are performed when new trajectory data arrives. S46. Based on the outlier distribution of normal trajectory samples within the sliding time window, calculate the local fraction threshold and use it as the current dynamic threshold. When the local outlier index of the target trajectory segment exceeds the dynamic threshold, generate an abnormal tendency signal for the corresponding trajectory segment. S47. Output the local outlier index of the trajectory segment and the corresponding anomaly tendency signal.

[0012] Optionally, step S5 includes: S51. Based on the airspace classification information of the monitoring area and aviation regulations, establish a flight physical envelope model and an airspace management rule base. The flight physical envelope model includes a speed-altitude envelope, a rate of climb limit curve, and a turning angular velocity limit. The airspace management rule base includes no-fly zones, altitude-restricted zones, and route boundary constraints. S52. For each trajectory segment, calculate the deviation metric from the flight physical envelope model, the deviation metric including velocity envelope exceeding the limit, climb rate exceeding the limit, and turn rate deviation. S53. Calculate the boundary crossing metric relative to the airspace management rule base, wherein the boundary crossing metric includes the proportion of time a trajectory segment enters the no-fly zone and the amount of altitude exceeding the limit within the altitude-restricted zone; S54. The deviation metric of the flight physical envelope model and the out-of-bounds metric of the airspace management rule base are weighted and combined to obtain the penalty term for the corresponding trajectory segment. The penalty term represents the degree of deviation of the trajectory segment at the level of physical constraints and airspace rules.

[0013] Optionally, step S6 includes: S61. Obtain the local outlier index, anomaly tendency signal, and penalty term; S62. Normalize the local outlier index and the penalty term to ensure that the range of values ​​for the local outlier index and the penalty term is consistent. S63. Construct a comprehensive anomaly scoring function based on preset weights, combine the normalized local outlier index with the penalty term in a weighted combination, and adjust the weighting coefficients based on the temporal continuity of the anomaly tendency signal to achieve a dynamic response to instantaneous anomaly trends and obtain a comprehensive anomaly score. S64. In the spatial grid and time grid, calculate the quantile threshold based on the comprehensive anomaly score distribution of normal samples in the near time window. When anomaly tendency signals are triggered continuously in adjacent time windows, reduce the corresponding quantile threshold to enhance the sensitivity of anomaly identification. S65. When the comprehensive anomaly score of the trajectory segment exceeds the quantile threshold, the trajectory segment is determined to be an abnormal flight behavior, the abnormal flight behavior identification result is output, and the corresponding spatial location and time interval are recorded.

[0014] Optionally, step S7 includes: S71. Extract the identifier, anomaly type, spatial location, and time interval information of the abnormal trajectory segment from the abnormal flight behavior identification results; S72. Generate a three-dimensional visual trajectory based on the spatial location of the abnormal trajectory segment, and form a spatial distribution map by combining the coordinates of the monitoring nodes; S73. The anomaly type, local outlier index, penalty item and comprehensive anomaly score of the abnormal trajectory segment are associated and stored to form an abnormal event data structure; S74. Generate an abnormal flight identification report, the report including abnormal trajectory segment identifier, abnormal type, abnormal time interval, abnormal score value and spatial location; S75. Store the abnormal flight identification report and related data in the monitoring database.

[0015] This invention addresses the issues of time asynchrony, coordinate system differences, and observation noise interference in low-altitude monitoring data acquisition by combining multi-source low-altitude monitoring data fusion with an improved TRAOD algorithm. It establishes a time reference using a unified timing signal and achieves high-precision synchronization and standardized preprocessing of multi-source monitoring data through sampling time interpolation and format standardization. In the trajectory modeling stage, it introduces a physical quantity-based distance measurement method based on planar coordinates, altitude, velocity, and heading angle. A distance matrix between trajectory segments is constructed through multi-dimensional feature weighted combination, effectively characterizing the comprehensive differences in spatial morphology and motion characteristics of flight trajectories. In the anomaly detection stage, the improved TRAOD algorithm incorporates a spatial index structure and local density. An incremental update mechanism combining outlier calculation and event anchor segmentation enables near real-time anomaly detection and dynamic threshold adjustment of trajectory data streams, outputting local outlier indicators and anomaly tendency signals. By combining a flight physical envelope model and airspace management rule base, penalties for exceeding speed envelope limits, climb rate limits, and outbound flight are calculated and weighted with the algorithm output to construct a comprehensive anomaly scoring function. Quantile adaptive thresholds enable accurate identification of anomalous trajectory segments. Finally, an anomaly flight identification report is generated, including anomaly type, spatial location, time interval, and anomaly score, establishing structured anomaly event data and achieving high-precision identification, visualization, and traceable management of anomalous flight behavior in low-altitude monitoring environments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall process of an intelligent identification method for abnormal flights based on low-altitude monitoring data proposed in this invention. Figure 2 This is a schematic diagram of the structure of the improved TRAOD algorithm in this invention; Figure 3 This is a schematic diagram of the abnormal comprehensive scoring calculation and judgment process in this invention; Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1-3 An intelligent identification method for abnormal flights based on low-altitude monitoring data includes the following steps: S1. Collect multi-source low-altitude monitoring data, and perform time base synchronization and data format unification processing on the multi-source low-altitude monitoring data to form a synchronized monitoring dataset. S2. Based on the flight event timestamp and preset time window, the synchronized monitoring dataset is divided into multiple sub-trajectory segments, and interpolation and smoothing are performed on each sub-trajectory segment. S3. Calculate the physical quantity sensing distance for any two sub-trajectory segments and perform standardization to form a physical quantity sensing distance matrix; S4. An improved TRAOD algorithm based on the physical quantity sensing distance matrix is ​​used to detect anomalies in each sub-trajectory segment and output local outlier index and anomaly tendency signal. S5. Calculate penalty terms for sub-trajectory segments based on flight physical envelope and airspace management rules; S6. Construct an abnormal comprehensive scoring function. Based on the abnormal tendency signal, combine the local outlier index and the penalty term according to a preset weight to obtain a comprehensive abnormal score. Calculate the quantile threshold. When the comprehensive abnormal score exceeds the corresponding quantile threshold, determine that the sub-trajectory segment is an abnormal flight behavior. S7. Generate an abnormal flight identification report and store the report in the monitoring database.

[0019] This invention first collects multi-source low-altitude monitoring data from radar, ADS-B, photoelectric sensors, and wireless direction-finding devices. Time synchronization is achieved via BeiDou or GPS timing signals, and a unified data format is used to form a synchronized monitoring dataset. The system divides trajectory segments based on flight event timestamps and preset time windows, generating standardized trajectory data through interpolation and smoothing. Subsequently, the physical quantity sensing distance between trajectory segments is calculated, and an improved TRAOD algorithm is input to perform anomaly detection, obtaining local outlier indices and anomaly tendency signals. Then, a penalty term is calculated by combining the flight physical envelope and airspace management rules to construct a comprehensive anomaly scoring function. Anomalies are determined for each trajectory segment, and an abnormal flight identification report is generated, completing the entire data analysis process.

[0020] In this embodiment, step S1 includes: S11. Deploy multi-source monitoring equipment in the low-altitude monitoring area. The multi-source monitoring equipment includes radar observation equipment, ADS-B receiver, photoelectric sensor and wireless direction finding device. Each device outputs the original observation data of the flying target according to the preset sampling frequency. S12. Collect the raw observation data and record the equipment identifier, flight event timestamp, monitoring node coordinates, target plane coordinates, altitude, heading angle, speed, signal strength, and data quality markers; S13. Establish a unified time reference, which is generated based on BeiDou or GPS timing signals and synchronizes each monitoring node through a network clock protocol, recording the synchronization status and clock offset. S14. Perform time base synchronization and use sampling time interpolation and alignment strategies to eliminate time drift; S15. Perform format unification processing, which includes field name standardization, measurement unit conversion, missing value placeholder filling and sensor type labeling. S16. The multi-source observation data that has been synchronized in time and formatted is aggregated into a synchronized monitoring dataset.

[0021] This invention deploys multi-source monitoring equipment within a low-altitude monitoring area, establishes a time reference through a unified time synchronization system, and uses a network clock protocol to achieve time synchronization across multiple nodes, recording synchronization status and offset information. During data acquisition, the system establishes original observation records based on device identification, timestamps, coordinates, and observation attributes, and performs format standardization and unit conversion on the outputs of various sensors. To avoid time drift caused by sampling clock errors from different monitoring sources, the system employs sampling time interpolation and alignment strategies for time synchronization correction. After time and format standardization, the multi-source data is fused to form a synchronized monitoring dataset, providing standardized input for subsequent trajectory segmentation and anomaly detection.

[0022] In this embodiment, step S2 includes: S21. The synchronized monitoring dataset is divided into multiple sub-trajectory segments based on the flight event timestamp and a preset time window. The preset time window is determined according to the target flight speed and sampling frequency. S22. Perform data sorting on each sub-trajectory segment after division, and arrange them in ascending order according to the flight event timestamp to form a time-continuous sequence of trajectory points; S23. Perform interpolation to fill missing time periods in the trajectory point sequence. The interpolation is based on the spatial coordinates, velocity and heading angle of adjacent valid trajectory points to calculate the position of the missing points. S24. Perform smoothing on the trajectory point sequence after interpolation and use a weighted average method within a sliding time window to suppress noise deviation. S25. Assign a unique identifier to each trajectory segment, and record the trajectory start and end times, number of samples, and monitoring source composition information, and store them as a standardized trajectory dataset.

[0023] In this invention, synchronized monitoring data is divided into trajectories based on flight event timestamps and preset time windows. The time windows are dynamically set according to the flight target speed range and sampling frequency to ensure that the number of data points within a trajectory segment meets the requirements for smoothing calculations. The system arranges the trajectory points within each segment in chronological order and uses interpolation based on the spatial position, velocity, and heading angle of adjacent valid trajectory points to fill sampling gaps. To reduce the impact of monitoring noise, the system applies a weighted average method for smoothing within the sliding time window, generating a continuous and stable sequence of trajectory points, forming a standardized trajectory dataset.

[0024] In this embodiment, step S3 includes: S31. For any two sub-trajectory segments in the standardized trajectory dataset, extract a set of trajectory points for distance calculation based on a uniform sampling interval. The set of trajectory points includes planar coordinates, altitude, velocity, and heading angle information. S32. Calculate the spatial path similarity between two sub-trajectory segments. The spatial path similarity is determined based on the sequence of changes in the plane coordinates of the trajectory points and the continuous trend of changes in the heading angle. S33. Calculate the differences between the two sub-track segments in terms of speed, altitude and heading angle, and obtain the average speed difference, average altitude difference and average heading angle difference respectively. S34. Based on the spatial path similarity, average speed difference, average altitude difference and average heading angle difference, set weighting coefficients and perform weighted combination to obtain the physical quantity sensing distance; S35. Perform standardization processing on the physical quantity sensing distance results to eliminate the dimensional differences of different physical quantities and form a comparable physical quantity sensing distance matrix.

[0025] In this invention, a set of trajectory points is extracted from any two sub-trajectory segments in a standardized trajectory dataset. This set includes planar coordinates, altitude, velocity, and heading angle information, and the number of trajectory points is ensured to be consistent based on a uniform sampling interval. The spatial path similarity between the two trajectory segments is calculated. This similarity is determined based on the sequence of changes in the planar coordinates of the trajectory points and the continuous trend of changes in the heading angle, and is numerically represented to characterize the similarity in trajectory shape and direction. Furthermore, the differences between the two trajectory segments in the dimensions of velocity, altitude, and heading angle are calculated, yielding the average velocity difference, average altitude difference, and average heading angle difference, respectively. Weighting coefficients are set according to these differences, and a weighted combination is performed to obtain the physical quantity sensing distance. Finally, the sensing distance results are zero-mean normalized and extreme value normalized to form a comparable distance matrix for subsequent anomaly detection.

[0026] In this embodiment, step S4 includes: S41. Anomaly detection is performed on the physical quantity sensing distance matrix using an improved TRAOD algorithm. The improved TRAOD algorithm includes four stages: spatial index construction, local density calculation, outlier calculation, and incremental update. S42. In the spatial index construction stage, a spatial index structure is established based on the distance relationship between trajectory segments in the physical quantity sensing distance matrix. The spatial index structure is constructed by combining grid partitioning and neighborhood radius search. S43. In the local density calculation stage, the local density value of the target trajectory segment is calculated based on the distance between the target trajectory segment and the corresponding trajectory segment in the physical quantity sensing distance matrix. S44. In the outlier calculation stage, the neighborhood density value is calculated. The neighborhood density value is the average value of the local density of each trajectory segment in the neighborhood set of the target trajectory segment. The local outlier index is determined based on the difference between the local density and the neighborhood density of the target trajectory segment. S45. In the incremental update phase, the continuous trajectory data is processed in batches based on the event anchor point segmentation structure. The event anchor point segmentation structure divides the trajectory into segments according to the event time anchor points set by the rate of change of heading angle, the rate of change of velocity, or the rate of change of altitude in the trajectory. Neighborhood incremental update and outlier correction are performed when new trajectory data arrives. The event anchor point segmentation structure automatically sets event time anchor points by detecting the rate of change of heading angle, velocity mutation rate, or altitude change rate in the trajectory, and divides the continuous trajectory into several trajectory segments using these anchor points. The improved TRAOD algorithm independently calculates local density and outlier indices within each segment and performs incremental updates based on the current segment when new trajectory data arrives. This structure effectively reduces the global recalculation overhead and achieves near real-time identification of abnormal low-altitude flight behavior; S46. Based on the outlier distribution of normal trajectory samples within the sliding time window, calculate the local fraction threshold and use it as the current dynamic threshold. When the local outlier index of the target trajectory segment exceeds the dynamic threshold, generate an abnormal tendency signal for the corresponding trajectory segment. The sliding time window is set according to the data update frequency of the monitoring system, preferably using a scrolling window mode. Each time new data is added, the window slides forward by a fixed time step, deleting the oldest sample and adding the newest sample. The system statistically analyzes the outlier distribution of normal samples within the window in real time and uses the quantile method to determine the threshold, ensuring that the anomaly judgment criteria are adaptively adjusted over time. The local quantile threshold is used for dynamic anomaly triggering control within the improved TRAOD algorithm. It is calculated independently of the quantile threshold in the comprehensive anomaly scoring stage and is used to adjust the anomaly detection sensitivity in real time during the trajectory flow input stage. The abnormal tendency signal refers to the binary judgment result generated by comparing the local outlier index calculated by the improved TRAOD algorithm with the dynamic threshold. It is used to indicate whether the trajectory segment has an abnormal trend. Its value can be 0 or 1, which respectively represent normal or have an abnormal tendency. S47. Output the local outlier index of the trajectory segment and the corresponding anomaly tendency signal.

[0027] In this invention, the physical quantity sensing distance matrix is ​​formed by weighting the spatial path similarity, average velocity difference, average altitude difference, and average heading angle difference between trajectory segments, and is used to characterize multidimensional physical differences. The improved TRAOD algorithm uses this matrix as input, first calculating the local density value of the trajectory segment within a preset neighborhood radius, and then determining the local outlier index based on the average density within the neighborhood. The system dynamically calculates the outlier distribution quantile of normal samples based on a sliding time window, obtaining a dynamically adjusted threshold in real time. When the outlier index of the target trajectory segment exceeds the threshold, an anomaly tendency signal is generated. A segmented structure based on event anchor points enables batch updates and neighborhood correction of trajectory data, thereby maintaining the real-time performance and accuracy of anomaly detection.

[0028] In this embodiment, step S5 includes: S51. Based on the airspace classification information of the monitoring area and aviation regulations, establish a flight physical envelope model and an airspace management rule base. The flight physical envelope model includes a speed-altitude envelope, a rate of climb limit curve, and a turning angular velocity limit. The airspace management rule base includes no-fly zones, altitude-restricted zones, and route boundary constraints. The flight physical envelope model is established based on aircraft type and flight performance parameters, and is obtained through statistical fitting of speed, altitude, and attitude data from historical flight samples. The flight physical envelope model can dynamically adjust parameters according to different aircraft types or load conditions to reflect the safe operating boundaries of the aircraft under different operating conditions. The airspace management rule base is generated based on airspace delineation information, geographic information system data, and regional flight restriction notices issued by the civil aviation authorities. The rule base stores airspace boundaries, altitude restrictions, and no-fly zones in geographic coordinates, and supports regular synchronous updates to maintain data validity. S52. For each trajectory segment, calculate the deviation metric from the flight physical envelope model, the deviation metric including velocity envelope exceeding the limit, climb rate exceeding the limit, and turn rate deviation. S53. Calculate the boundary crossing metric relative to the airspace management rule base, wherein the boundary crossing metric includes the proportion of time a trajectory segment enters the no-fly zone and the amount of altitude exceeding the limit within the altitude-restricted zone; S54. The deviation metric of the flight physical envelope model and the out-of-bounds metric of the airspace management rule base are weighted and combined to obtain the penalty term for the corresponding trajectory segment. The penalty term represents the degree of deviation of the trajectory segment at the level of physical constraints and airspace rules.

[0029] The weighted combination employs a linear weighted model, with weight coefficients determined based on the statistical importance of envelope deviation and boundary crossing metrics. Initial weights are set through correlation analysis of historical normal trajectory samples in the initial stage, and are adaptively adjusted based on error feedback from anomaly detection results during system operation. The final penalty term's numerical range is normalized and used for anomaly scoring correction.

[0030] In this embodiment, step S6 includes: S61. Obtain the local outlier index, anomaly tendency signal, and penalty term; S62. Normalize the local outlier index and the penalty term to ensure that the range of values ​​for the local outlier index and the penalty term is consistent. S63. Construct a comprehensive anomaly scoring function based on preset weights, combine the normalized local outlier index with the penalty term in a weighted combination, and adjust the weighting coefficients based on the temporal continuity of the anomaly tendency signal to achieve a dynamic response to instantaneous anomaly trends and obtain a comprehensive anomaly score. The comprehensive anomaly scoring function is constructed using a linear weighted model. The weights are determined based on the contribution of the local outlier index and the penalty term to historical anomaly identification. The initial weights are obtained by training to minimize the scoring error of historical samples and are adaptively updated based on the identification accuracy during system operation. S64. In the spatial grid and time grid, calculate the quantile threshold based on the comprehensive anomaly score distribution of normal samples in the near time window. When anomaly tendency signals are triggered continuously in adjacent time windows, reduce the corresponding quantile threshold to enhance the sensitivity of anomaly identification. The spatial grid is a discrete spatial unit that divides the three-dimensional space of the monitoring area according to the dimensions of longitude, latitude, and altitude. It is used to statistically analyze the distribution of abnormal scores for trajectory segments within the spatial range. The temporal grid is a fixed time window that divides the monitoring period. It is used to record the statistical results of abnormal scores within the corresponding time period. The system constructs a joint distribution matrix based on the spatial grid and the temporal grid, and calculates the comprehensive score quantile threshold of normal samples according to region and time period, respectively, so as to realize the adaptive adjustment of the abnormal identification threshold in the spatiotemporal dimensions. The near-time window normal samples are obtained by statistically analyzing trajectory segments with outliers below a fixed empirical threshold within a sliding time window. It is preferable to take the sample data from the most recent N time windows to ensure the timeliness and representativeness of the threshold calculation. The continuous triggering determination of the abnormal tendency signal is based on a fixed number of sliding time windows. Preferably, when the signal value is in the triggering state within three to five consecutive time windows, it is considered as a continuous abnormal trend, which is used to reduce the determination threshold and improve the recognition sensitivity. S65. When the comprehensive anomaly score of the trajectory segment exceeds the quantile threshold, the trajectory segment is determined to be an abnormal flight behavior, the abnormal flight behavior identification result is output, and the corresponding spatial location and time interval are recorded.

[0031] In this embodiment, step S7 includes: S71. Extract the identifier, anomaly type, spatial location, and time interval information of the abnormal trajectory segment from the abnormal flight behavior identification results; The anomaly type is determined based on a combination of the detection results from the improved TRAOD algorithm and the flight physical envelope model. When performing local outlier analysis, the system determines the anomaly pattern based on the contribution ratio of each dimension—speed, altitude, heading angle, and spatial position—to the outlier index. When the contribution rate of a certain dimension exceeds a preset threshold, the corresponding type of statistical anomaly is marked. Simultaneously, the system determines physical constraint anomalies based on the flight physical envelope model and the types of penalty items triggered in the airspace management rule base. Finally, the system merges the two types of results to generate anomaly type labels including categories such as spatial deviation, sudden speed changes, heading anomalies, envelope violations, and outbound flight, which are used for generating abnormal flight reports. S72. Generate a three-dimensional visual trajectory based on the spatial location of the abnormal trajectory segment, and form a spatial distribution map by combining the coordinates of the monitoring nodes; S73. The anomaly type, local outlier index, penalty item and comprehensive anomaly score of the abnormal trajectory segment are associated and stored to form an abnormal event data structure; S74. Generate an abnormal flight identification report, the report including abnormal trajectory segment identifier, abnormal type, abnormal time interval, abnormal score value and spatial location; S75. Store the abnormal flight identification report and related data in the monitoring database. Example

[0032] To verify the feasibility and effectiveness of this invention in actual low-altitude monitoring scenarios, it was applied to a low-altitude flight monitoring and management platform in a coastal city. This city has a relatively complete drone monitoring system, primarily monitoring logistics drones, aerial photography drones, and municipal inspection drones. Both fixed-route flights and temporary mission flights occur simultaneously in the airspace environment. The monitoring area is approximately 120 square kilometers, covering the urban area, port, and suburban transition zone. The system is equipped with 6 ADS-B receivers, 4 X-band radars, and 2 sets of optoelectronic observation equipment. Each monitoring node is synchronized via the BeiDou time synchronization network, with a sampling frequency between 1 and 5 Hz, forming a multi-source heterogeneous monitoring data stream. Under existing technological conditions, common problems include inconsistent time bases for multi-source data, large differences in observation accuracy between different monitoring sources, frequent track breaks and noise drift, and high anomaly identification delays, making it difficult to meet the real-time requirements of urban low-altitude safety management.

[0033] In this embodiment, precise synchronization of multi-source data is first achieved by establishing a unified time reference. The BeiDou time signal serves as the unified reference for the entire network, and node synchronization is performed via the Network Clock Protocol (NTP), maintaining a time offset within 2 milliseconds. Observational data output from various sensors undergo field unification, coordinate system transformation, and missing data completion to generate a synchronized monitoring dataset. Subsequently, the data is segmented into trajectory segments based on flight event timestamps and a 10-second preset time window. Weighted interpolation is used to fill in missing data points, and the trajectory points are smoothed, improving trajectory continuity by approximately 18.1% and reducing the average trajectory deviation from 1.20 meters to 0.83 meters.

[0034] In the anomaly detection stage, a physical quantity sensing distance matrix is ​​constructed based on the planar coordinates, altitude, velocity, and heading angle of the trajectory segments. The improved TRAOD algorithm proposed in this invention is then used for trajectory-level anomaly identification. This algorithm establishes a spatial index structure by combining grid partitioning with neighborhood radius search, improving neighborhood search efficiency by approximately 41% compared to the traditional KD tree. An event anchor point segmentation structure is introduced in the local density and outlier calculation stage, automatically dividing event segments based on the heading angle change rate, velocity mutation rate, and altitude change rate, thereby improving the sensitivity of local features to anomalies. The system performs detection on approximately 42,000 trajectory data points in a GPU parallel environment, generating 198,000 sub-trajectory segments with an average detection latency of 0.89 seconds, achieving near real-time anomaly identification.

[0035] Furthermore, penalty terms are calculated based on the flight physical envelope model of the monitored area and the airspace management rule base. The physical envelope model sets a maximum speed of 60 m / s, a maximum rate of climb of 8 m / s, and a maximum turn rate of 15° / s. The airspace rule base includes three no-fly zones and two altitude-restricted zones, with boundary data sourced from local air traffic control and civil aviation management documents. For trajectory segments exhibiting speed exceeding limits, excessive climb, or boundary violations, the system automatically calculates penalty terms and weights them with local outlier indicators to generate a comprehensive anomaly scoring function. The comprehensive score dynamically calculates quantile thresholds based on the distribution of normal samples within the near-time window. When the score exceeds the 95th quantile threshold, the segment is determined to be an abnormal flight.

[0036] To verify the beneficial effects of the present invention, the performance of the method of the present invention and the traditional trajectory anomaly detection algorithm were compared on the same dataset. The experimental results are shown in Table 1: Table 1. Comparison of Anomaly Detection Performance of Different Algorithms in Low-Altitude Monitoring Scenarios Comparison indicators Method of the present invention Traditional methods Average detection delay (s) 0.89 1.83 Anomaly detection accuracy (%) 95.8 88.6 False alarm rate (%) 3.7 7.5 Trajectory continuity error (m) 0.83 1.35 As shown in Table 1, the improved TRAOD algorithm of this invention outperforms traditional algorithms in terms of detection accuracy, real-time performance, and trajectory continuity. It is particularly stable under multi-source monitoring noise conditions, and its robustness in complex low-altitude environments is significantly improved. This invention detected a total of 117 abnormal flight events, including 37 cases of unauthorized climbs, 19 no-fly zone incursions, 26 cases of sudden circling, and 35 cases of low-altitude high-speed flight. All events were manually verified, achieving an accuracy rate of 95.8% and a false alarm rate of 3.7%. Compared to traditional algorithms, this represents an 8 percentage point improvement in accuracy and a 5 percentage point reduction in false alarm rate. The average report generation time is 2.9 seconds, approximately 58% lower than existing systems, significantly improving the real-time response efficiency for abnormal events. Through multi-source synchronization, physical quantity sensing modeling, and dynamic quantile threshold determination mechanisms, this invention achieves closed-loop processing of low-altitude abnormal flight behavior from acquisition and identification to report generation. It possesses real-time, high-precision, and interpretable technical advantages, providing reliable data support and algorithmic foundation for urban low-altitude monitoring and safety management.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of abnormal flights based on low-altitude monitoring data, characterized in that, Includes the following steps: S1. Collect multi-source low-altitude monitoring data, and perform time base synchronization and data format unification processing on the multi-source low-altitude monitoring data to form a synchronized monitoring dataset. S2. Based on the flight event timestamp and preset time window, the synchronized monitoring dataset is divided into multiple sub-trajectory segments, and interpolation and smoothing are performed on each sub-trajectory segment. S3. Calculate the physical quantity sensing distance for any two sub-trajectory segments and perform standardization to form a physical quantity sensing distance matrix; S4. An improved TRAOD algorithm based on the physical quantity sensing distance matrix is ​​used to detect anomalies in each sub-trajectory segment and output local outlier index and anomaly tendency signal. S5. Calculate penalty terms for sub-trajectory segments based on flight physical envelope and airspace management rules; S6. Construct an abnormal comprehensive scoring function. Based on the abnormal tendency signal, combine the local outlier index and the penalty term according to a preset weight to obtain a comprehensive abnormal score. Calculate the quantile threshold. When the comprehensive abnormal score exceeds the corresponding quantile threshold, determine that the sub-trajectory segment is an abnormal flight behavior. S7. Generate an abnormal flight identification report and store the abnormal flight identification report in the monitoring database.

2. The method for intelligent identification of abnormal flights based on low-altitude monitoring data according to claim 1, characterized in that, Step S1 includes: S11. Deploy multi-source monitoring equipment in the low-altitude monitoring area. The multi-source monitoring equipment includes radar observation equipment, ADS-B receiver, photoelectric sensor and wireless direction finding device. Each device outputs the original observation data of the flying target according to the preset sampling frequency. S12. Collect the raw observation data and record the equipment identifier, flight event timestamp, monitoring node coordinates, target plane coordinates, altitude, heading angle, speed, signal strength, and data quality markers; S13. Establish a unified time reference, which is generated based on BeiDou or GPS timing signals and synchronizes each monitoring node through a network clock protocol, recording the synchronization status and clock offset. S14. Perform time base synchronization and use sampling time interpolation and alignment strategies to eliminate time drift; S15. Perform format unification processing, which includes field name standardization, measurement unit conversion, missing value placeholder filling and sensor type labeling. S16. The multi-source observation data that has been synchronized in time and formatted is aggregated into a synchronized monitoring dataset.

3. The method for intelligent identification of abnormal flights based on low-altitude monitoring data according to claim 1, characterized in that, Step S2 includes: S21. The synchronized monitoring dataset is divided into multiple sub-trajectory segments based on the flight event timestamp and a preset time window. The preset time window is determined according to the target flight speed and sampling frequency. S22. Perform data sorting on each sub-trajectory segment after division, and arrange them in ascending order according to the flight event timestamp to form a time-continuous sequence of trajectory points; S23. Perform interpolation to fill missing time periods in the trajectory point sequence. The interpolation is based on the spatial coordinates, velocity and heading angle of adjacent valid trajectory points to calculate the position of the missing points. S24. Perform smoothing on the trajectory point sequence after interpolation and use a weighted average method within a sliding time window to suppress noise deviation. S25. Assign a unique identifier to each trajectory segment, and record the trajectory start and end times, number of samples, and monitoring source composition information, and store them as a standardized trajectory dataset.

4. The method for intelligent identification of abnormal flights based on low-altitude monitoring data according to claim 3, characterized in that, Step S3 includes: S31. For any two sub-trajectory segments in the standardized trajectory dataset, extract a set of trajectory points for distance calculation based on a uniform sampling interval. The set of trajectory points includes planar coordinates, altitude, velocity, and heading angle information. S32. Calculate the spatial path similarity between two sub-trajectory segments. The spatial path similarity is determined based on the sequence of changes in the plane coordinates of the trajectory points and the continuous trend of changes in the heading angle. S33. Calculate the differences between the two sub-track segments in terms of speed, altitude and heading angle, and obtain the average speed difference, average altitude difference and average heading angle difference respectively. S34. Based on the spatial path similarity, average speed difference, average altitude difference and average heading angle difference, set weighting coefficients and perform weighted combination to obtain the physical quantity sensing distance; S35. Perform standardization processing on the physical quantity sensing distance results to eliminate the dimensional differences of different physical quantities and form a comparable physical quantity sensing distance matrix.

5. The method for intelligent identification of abnormal flight based on low-altitude monitoring data according to claim 4, characterized in that, Step S4 includes: S41. Anomaly detection is performed on the physical quantity sensing distance matrix using an improved TRAOD algorithm. The improved TRAOD algorithm includes four stages: spatial index construction, local density calculation, outlier calculation, and incremental update. S42. In the spatial index construction stage, a spatial index structure is established based on the distance relationship between trajectory segments in the physical quantity sensing distance matrix. The spatial index structure is constructed by combining grid partitioning and neighborhood radius search. S43. In the local density calculation stage, the local density value of the target trajectory segment is calculated based on the distance between the target trajectory segment and the corresponding trajectory segment in the physical quantity sensing distance matrix. S44. In the outlier calculation stage, the neighborhood density value is calculated. The neighborhood density value is the average value of the local density of each trajectory segment in the neighborhood set of the target trajectory segment. The local outlier index is determined based on the difference between the local density and the neighborhood density of the target trajectory segment. S45. In the incremental update phase, the continuous trajectory data is processed in batches based on the event anchor point segmentation structure. The event anchor point segmentation structure divides the trajectory into segments according to the event time anchor points set by the rate of change of heading angle, the rate of change of velocity, or the rate of change of altitude in the trajectory. Neighborhood incremental update and outlier correction are performed when new trajectory data arrives. S46. Based on the outlier distribution of normal trajectory samples within the sliding time window, calculate the local fraction threshold and use it as the current dynamic threshold. When the local outlier index of the target trajectory segment exceeds the dynamic threshold, generate an abnormal tendency signal for the corresponding trajectory segment. S47. Output the local outlier index of the trajectory segment and the corresponding anomaly tendency signal.

6. The method for intelligent identification of abnormal flights based on low-altitude monitoring data according to claim 1, characterized in that, Step S5 includes: S51. Based on the airspace classification information of the monitoring area and aviation regulations, establish a flight physical envelope model and an airspace management rule base. The flight physical envelope model includes a speed-altitude envelope, a rate of climb limit curve, and a turning angular velocity limit. The airspace management rule base includes no-fly zones, altitude-restricted zones, and route boundary constraints. S52. For each trajectory segment, calculate the deviation metric from the flight physical envelope model, the deviation metric including velocity envelope exceeding the limit, climb rate exceeding the limit, and turn rate deviation. S53. Calculate the boundary crossing metric relative to the airspace management rule base, wherein the boundary crossing metric includes the proportion of time a trajectory segment enters the no-fly zone and the amount of altitude exceeding the limit within the altitude-restricted zone; S54. The deviation metric of the flight physical envelope model and the out-of-bounds metric of the airspace management rule base are weighted and combined to obtain the penalty term for the corresponding trajectory segment. The penalty term represents the degree of deviation of the trajectory segment at the level of physical constraints and airspace rules.

7. The method for intelligent identification of abnormal flights based on low-altitude monitoring data according to claim 5, characterized in that, Step S6 includes: S61. Obtain the local outlier index, anomaly tendency signal, and penalty term; S62. Normalize the local outlier index and the penalty term to ensure that the range of values ​​for the local outlier index and the penalty term is consistent. S63. Construct a comprehensive anomaly scoring function based on preset weights, combine the normalized local outlier index with the penalty term in a weighted combination, and adjust the weighting coefficients based on the temporal continuity of the anomaly tendency signal to achieve a dynamic response to instantaneous anomaly trends and obtain a comprehensive anomaly score. S64. In the spatial grid and time grid, calculate the quantile threshold based on the comprehensive anomaly score distribution of normal samples in the near time window. When anomaly tendency signals are triggered continuously in adjacent time windows, reduce the corresponding quantile threshold to enhance the sensitivity of anomaly identification. S65. When the comprehensive anomaly score of the trajectory segment exceeds the quantile threshold, the trajectory segment is determined to be an abnormal flight behavior, the abnormal flight behavior identification result is output, and the corresponding spatial location and time interval are recorded.

8. The method for intelligent identification of abnormal flight based on low-altitude monitoring data according to claim 7, characterized in that, Step S7 includes: S71. Extract the identifier, anomaly type, spatial location, and time interval information of the abnormal trajectory segment from the abnormal flight behavior identification results; S72. Generate a three-dimensional visual trajectory based on the spatial location of the abnormal trajectory segment, and form a spatial distribution map by combining the coordinates of the monitoring nodes; S73. The anomaly type, local outlier index, penalty item and comprehensive anomaly score of the abnormal trajectory segment are associated and stored to form an abnormal event data structure; S74. Generate an abnormal flight identification report, the report including abnormal trajectory segment identifier, abnormal type, abnormal time interval, abnormal score value and spatial location; S75. Store the abnormal flight identification report and related data in the monitoring database.

Images