A flight consistency warning method and system based on multi-dimensional score fusion and intelligent grading mechanism

Abstract

The application discloses a flight consistency alarm method and system based on multi-dimensional score fusion and intelligent grading mechanism. The method comprises the following steps: obtaining a comprehensive consistency score for evaluating the consistency between an actual flight track of an aircraft and a flight plan track according to multi-dimensional scores; the multi-dimensional score results comprise a horizontal distance score, a vertical height score, a heading angle score and a dynamic time warping track shape alignment score; judging the level of deviation between the actual flight track and the flight plan track based on the comprehensive consistency score, and outputting alarm information; the alarm information comprises an alarm level. The application can timely and accurately reflect the consistency deviation between the flight track and the flight plan.

Description

Technical Field

[0001] This application relates to the field of airspace safety technology, and in particular to a flight consistency warning method and system based on multi-dimensional scoring fusion and intelligent classification mechanism. Background Technology

[0002] Flight consistency monitoring is one of the fundamental technologies for ensuring flight safety in air traffic management. Currently, various civil aviation radar systems and ADS-B data systems can provide high-frequency flight track data, but air traffic control systems often rely on static rules (such as position deviation, track deviation distance, etc.) for judgment in their alarm modules.

[0003] With increasing airspace complexity and more low-altitude / UAM operations, traditional methods are struggling to accurately and promptly reflect the degree of consistency deviation between flight tracks and flight plans, and there is a lack of intelligent alarm assessment models.

[0004] Therefore, existing technologies lack an intelligent track consistency alarm system that integrates track morphology analysis, scoring function modeling, level classification, parameter adjustment, and adaptive optimization.

[0005] Existing technologies lack effective modeling of the overall shape of flight paths. Existing methods rely heavily on single-point deviation judgments at location points, which cannot accurately capture morphological differences such as flight path trends and curvature changes, leading to missed or false alarms. The alarm logic of existing technologies is simple and cannot express the severity level of the deviation. Most existing systems are binary alarms (consistent or inconsistent), lacking continuous scoring and multi-level classification mechanisms, which makes it difficult to meet the needs of refined control in complex airspace. Existing technologies lack flexible scoring strategy configuration and adaptive mechanisms. The parameters in existing alarm modules are fixed and cannot be dynamically adjusted according to flight scenarios, stages, or operational data, which affects the versatility of the system and the actual deployment efficiency. Summary of the Invention

[0006] In view of this, this application provides a flight consistency alarm method and system based on multi-dimensional scoring fusion and intelligent grading mechanism, which aims to measure the degree of consistency between flight track and flight plan from multiple dimensions, and combine a configurable scoring mechanism and intelligent alarm grading mechanism to generate interpretable and controllable flight deviation alarm results, which can reflect the degree of consistency deviation between flight track and flight plan in a timely and accurate manner.

[0007] This application discloses a flight consistency alarm method based on multi-dimensional score fusion and intelligent classification mechanism, which includes: Step 1: Based on the multi-dimensional scores, obtain a comprehensive consistency score to evaluate the consistency between the actual flight path and the planned flight path of the aircraft; the multi-dimensional scores include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped path alignment score. Step 2: Based on the comprehensive consistency score, determine the degree of deviation between the actual flight track and the planned flight track, and output alarm information; the alarm information includes the alarm level.

[0008] Further, step 1 includes: Step 11: Calculate the horizontal distance score: Synchronize and match the actual flight track with the planned flight track to ensure a one-to-one correspondence between corresponding time points; calculate the geographical distance error under the horizontal projection; normalize all distance differences to obtain the horizontal distance score. Where 1 represents complete agreement and 0 represents maximum deviation; Step 12: Calculate the vertical altitude score: Extract actual flight altitude and planned altitude data and perform time synchronization matching; calculate the altitude difference between each pair of matching points; normalize the data according to the set maximum allowable altitude deviation to obtain the vertical altitude score. ; Step 13: Calculate the heading angle score: Extract the actual heading angle and the planned heading angle data and perform time synchronization matching; calculate the heading angle difference between each pair of matching points, with a value range of [0°, 180°]; normalize the angle difference to obtain the heading angle score. ; Step 14: Introduce the Dynamic Time Warping (DTW) algorithm to perform temporal spatial similarity alignment calculation between the flight plan track and the real-time flight track, and obtain the dynamic time warping track morphology alignment score; Step 15: Based on the horizontal distance score, vertical height score, heading angle score, and dynamic time-warped track alignment score obtained in Steps 11 to 14, a comprehensive consistency score is obtained.

[0009] Further, step 14 includes: Step 141: The system first processes the input flight plan track. and real-time flight track To be processed uniformly; This is the nth waypoint of the planned flight path. For the m-th trackpoint of the real-time flight track, coordinate sampling and interpolation are performed on the flight plan track and the real-time flight track to make the number of trackpoints approximately equal; the format of each trackpoint is unified to a three-dimensional geographic location. , All waypoints have undergone coordinate transformation for spatial distance calculation. , , These are the longitude, latitude, and altitude of the i-th track point on the flight plan track, respectively. , , These are the longitude, latitude, and altitude of the j-th trackpoint on the real-time flight path, respectively. Step 142: Define the following three-dimensional spatial distance function to represent track point pairs. Distance error between: ,in, This represents the spatial difference in the x-axis, y-axis, and z-axis directions after coordinate system transformation; Step 143: Construct a two-dimensional distance matrix Each of its elements is The minimum cumulative path cost is calculated using a dynamic programming algorithm.

[0010] in, To satisfy the alignment constraints of an effective path, The total cost of the path between the planned flight track and the real-time flight track; Step 144: Calculate the total cost of the path using the following formula. With the set maximum acceptable alignment distance Normalization is performed to obtain the dynamic time-warped track morphology alignment score. :

[0011] The smaller the cost of track alignment, the closer the score is to 1, indicating that the overall shape of the tracks is more similar; if the cost exceeds the set tolerance, the score will approach 0.

[0012] Further, step 15 includes: The scores from each dimension are weighted and fused together to obtain a comprehensive consistency score using the following formula, which serves as the basis for alarm determination:

[0013] in, To achieve a comprehensive consistency score, the weights for the horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped track alignment score are as follows: And satisfy ; The higher the value, the better the consistency of the flight path.

[0014] Furthermore, it also includes a scoring function configuration mechanism; the scoring function configuration mechanism is used to dynamically adjust key parameters and supports runtime hot updates without restarting the main system; the key parameters include the weights, window lengths, and threshold ranges of various track scores; The scoring function consists of multiple sub-scoring items, and the system supports the definition and updating of the following parameters: Weights of horizontal distance score, vertical altitude score, heading angle score, and dynamic time-normalized track alignment score; scoring interval boundaries used for alarm level determination; maximum tolerance limit for DTW distance; sliding window time range; altitude error tolerance threshold; heading error tolerance threshold; During startup, the trajectory consistency scoring module loads the configuration file in the specified path and establishes a memory mapping. All subsequent scoring calculations read the memory parameters formed by the configuration file, eliminating the need for hard coding. The trajectory consistency scoring module uses a configuration manager, which is responsible for periodically monitoring changes to the configuration file and reloading the parameters when changes occur. It provides a read-only interface for the trajectory consistency scoring module to call. The track consistency scoring module implements a hot update mechanism: a file monitoring mechanism; asynchronous loading is triggered after a change in file content is detected; automatic rollback to the previous version is performed when loading fails; and configuration can be uploaded via a web interface.

[0015] Further, step 2 includes: Step 21: The alarm level determination module receives the comprehensive consistency score sent by the track consistency scoring module and obtains the average speed of the current track segment; Step 22: Determine whether to use the decision tree module. If so, load the decision tree module, perform model inference, and output the alarm level. Otherwise, use a lightweight fine-tuning mechanism for the scoring threshold to perform scoring trend analysis and boundary stability judgment to obtain the alarm level threshold. Compare the overall consistency score with the alarm level threshold and output the alarm level.

[0016] Further, in step 22, the decision tree module is loaded to perform model inference and output alarm levels, including: Training phase: The training set is built offline, with the comprehensive consistency score as the input and the consistency level of historical manual assessment as the label. The CART classification tree model is used for fitting, and the tree model structure is output. Deployment phase: After the model is loaded, the score results of each sliding window are judged in real time. The system supports calling the interface to complete the classification. The judgment result and the score value are output together for use by the subsequent alarm output optimization module. The system supports returning the current track score input, the hit tree branch path, and the level output as structured JSON. In step 22, the input data for the lightweight fine-tuning mechanism of the scoring threshold is: the real-time consistency score value score_t; the scoring sequence S=[score_{t} within the sliding time window. n},…,score_{t 1}], n is the nth time window; the initial alarm level threshold set T={τ_1,τ_2,…,τ_k} is set by the system; lightweight fine-tuning parameters are provided through the configuration file; lightweight fine-tuning parameters include floating threshold factor and jump suppression window length; Scoring trend analysis and boundary stability judgment: In each alarm cycle, the system calculates the statistical characteristics of the scoring sliding window, including the mean μ, standard deviation σ, and alarm level jump frequency; if it is found that the score is still near the boundary of a certain alarm level after a preset time, or the level fluctuates frequently, it is determined that the current threshold setting cannot better distinguish the actual level status. If the current threshold setting cannot better distinguish the actual level status, the output of the scoring trend analysis and boundary stability judgment will be fine-tuned. The lightweight fine-tuning mechanism for the scoring threshold outputs the adjusted alarm level threshold. This information is used by the subsequent alarm output optimization module and can output diagnostic information for the log system to evaluate. The diagnostic information includes fine-tuning direction, oscillation intensity, and trigger flags. This is the threshold for the kth alarm level.

[0017] Furthermore, the threshold is fine-tuned using the following formula:

[0018] in, The alarm level threshold before adjustment The adjusted alarm level threshold. To fine-tune the scaling factor, The standard deviation of the scoring sequence; This indicates a slight upward or downward adjustment; the direction of the adjustment is determined by the direction of the score's offset at the boundary; if the score tends to stabilize, the threshold will return to the original set value.

[0019] Furthermore, after step 2, step 3 is also included; step 3 includes: the alarm output optimization module improves the accuracy of alarm output and user experience through a multi-level alarm output optimization mechanism; the multi-level alarm output optimization mechanism includes alarm deduplication, time aggregation mechanism, alarm level suppression strategy and sliding smoothing mechanism; Alarm deduplication: For the same flight target, the system generates a consistency score and level judgment in each evaluation window; if multiple consecutive windows generate alarms of the same level, the system will only output the alarm information once to avoid repeated push of the same alarm; the deduplication judgment conditions include the unique identifier of the aircraft, the alarm level is consistent in adjacent time windows, and the time interval is less than the set deduplication threshold. Time aggregation mechanism: The system introduces a time aggregation window to aggregate multiple alarm records within a short period of time into a single high-priority alarm; the aggregation logic is as follows: set an aggregation window, count relevant indicators within the window, and output aggregated alarms; relevant indicators include alarm count, maximum level, and average score; aggregated alarms include start time, duration, maximum level, and average deviation score; Alarm level suppression strategy: The system is designed with an alarm level lag judgment mechanism. The judgment mechanism is as follows: if the alarm level decreases in the current window compared to the previous window, the alarm will only be officially reduced after two consecutive windows of decrease; if the level increases, a new alarm will be output immediately. Sliding smoothing mechanism: The scoring trend is averaged to suppress scoring fluctuations; the smoothed score is used as a reference feature to determine the alarm level, and thus weights are assigned. The output of the alarm output optimization module includes alarm ID, aircraft ID, timestamp, as well as level, score, alarm triggering rule, and alarm type description.

[0020] This application also discloses a flight consistency alarm system based on multi-dimensional score fusion and intelligent grading mechanism, which implements the flight consistency alarm method based on multi-dimensional score fusion and intelligent grading mechanism described above, comprising: The trajectory consistency scoring module is used to obtain a comprehensive consistency score based on multi-dimensional scoring to evaluate the consistency between the actual flight trajectory and the planned flight trajectory of the aircraft. The multi-dimensional scoring results include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-curved trajectory alignment score. The alarm level determination module is used to determine the degree of deviation between the actual flight track and the flight plan track based on the comprehensive consistency score, and output alarm information; the alarm information includes the alarm level.

[0021] This application significantly improves the performance and applicability of the flight consistency alarm system by constructing a multi-dimensional consistency scoring mechanism, an intelligent alarm classification model, a parameter configuration system, and an adaptive scoring threshold mechanism. Specific beneficial effects include: 1. More accurate consistency assessment: The introduction of the DTW algorithm to model the overall shape of the flight track can more accurately capture the trend deviation of the track compared with the traditional point-to-point deviation method.

[0022] 2. Supports multi-dimensional score fusion to reduce false alarms and missed alarms: The four-dimensional scores of horizontal, vertical, heading, and DTW constitute a more comprehensive deviation perception system. Through configurable weighted calculation, the coverage and robustness of alarm judgment are improved.

[0023] 3. Alarm levels are hierarchical and interpretable: The system no longer uses traditional binary judgment, but outputs hierarchical alarm levels (normal, minor, moderate, severe), which helps controllers to make differentiated responses.

[0024] 4. Supports both model-driven and parameter-driven decision-making: The system supports decision-making through trained decision tree models, and can also use scoring thresholds for decision-making in scenarios where the model is missing or in a cold start, demonstrating strong engineering deployability.

[0025] 5. The scoring threshold fine-tuning mechanism is highly adaptive: even without historical samples, it can still dynamically optimize the level judgment boundary based on the scoring fluctuation pattern, effectively suppressing alarm oscillations and misjudgments.

[0026] 6. Modular architecture with support for hot configuration file updates: Parameters for each module support YAML / JSON format configuration and have online update capabilities, enabling rapid adaptation to different airport, airline, and airspace policies. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0028] Figure 1 This is a flowchart illustrating a flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to an embodiment of this application. Figure 2 This is a schematic diagram of the data processing of the track consistency scoring module in an embodiment of this application; Figure 3 This is a schematic diagram of the data processing of the alarm level determination module in an embodiment of this application. Detailed Implementation

[0029] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.

[0030] This application provides a flight consistency alarm method and system based on multi-dimensional scoring fusion and intelligent grading mechanisms. It aims to measure the degree of consistency between flight trajectory and flight plan from multiple dimensions, and combines a configurable scoring mechanism and an intelligent alarm grading mechanism to generate interpretable and controllable flight deviation alarm results. See also... Figure 1 This application proposes a flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism, which includes: Step 1: Based on the multi-dimensional scores, obtain a comprehensive consistency score to evaluate the consistency between the actual flight path and the planned flight path of the aircraft; the multi-dimensional scores include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped path alignment score. Step 2: Based on the comprehensive consistency score, determine the degree of deviation between the actual flight track and the planned flight track, and output alarm information; the alarm information includes the alarm level.

[0031] The implementation process of step 1 above includes: The trajectory consistency scoring module aims to comprehensively evaluate the consistency between the aircraft's actual flight trajectory and the planned flight path, providing multi-dimensional scoring results to support subsequent alarm determination and risk assessment. The flowchart of the trajectory consistency scoring module is shown below. Figure 2 As shown.

[0032] (1) Horizontal distance score: Horizontal distance rating measures the degree of deviation of an aircraft from its planned horizontal flight path. The horizontal distance rating process is as follows: The actual flight track is time-synchronized with the planned flight track to ensure a one-to-one correspondence between corresponding time points; the geographical distance error under horizontal projection is calculated; and all distance differences are normalized to obtain the horizontal distance score. , where 1 represents complete consistency and 0 represents maximum deviation.

[0033] (2) Vertical height score: Vertical altitude scoring assesses the degree of deviation of an aircraft from its planned vertical altitude. The vertical altitude scoring process is as follows: Extract actual flight altitude and planned altitude data and perform time synchronization matching. Calculate the altitude difference between each pair of matching points. Normalize the data based on the set maximum allowable altitude deviation (e.g., 300 meters) to obtain a vertical altitude score. .

[0034] (3) Heading angle rating: The heading angle rating measures the consistency between an aircraft's heading and its planned heading. The heading angle rating process is as follows: Extract the actual and planned heading angle data and perform time synchronization matching. Calculate the heading angle difference between each pair of matching points, with a value ranging from [0°, 180°]. Normalize the angle difference (divide by 180°) to obtain the heading angle score. .

[0035] (4) Dynamic time-normalized track alignment score: Traditional trajectory consistency assessment methods are mostly based on "point-to-point" errors and lack the ability to model the overall shape of the flight trajectory (such as deviation trends, turning patterns, delay regression, etc.), which makes it impossible to capture the differences in the global structure between the aircraft and the planned trajectory.

[0036] This application introduces a Dynamic Time Warping (DTW) algorithm to perform temporal spatial similarity calculation between the flight plan track and the real-time flight track, thereby achieving a quantitative score of track morphology consistency without forcing "isochronous matching".

[0037] DTW Algorithm Principles and Application Scenarios: DTW is an algorithm for aligning two time series, allowing the sequences to be “compressed” or “stretched” on the time axis to minimize the cumulative distance between the two sequences.

[0038] Planned flight paths are represented as sequences. Each point Real-time flight tracks are represented as sequences. Each point Find a path This minimizes the total alignment cost between the two tracks:

[0039] in () represents the geographical distance between points.

[0040] The specific process of applying the DTW algorithm principle to flight track alignment includes track preprocessing and sampling standardization, and unifying the format of each track point into a three-dimensional geographic location. Track preprocessing and sampling standardization: The system first processes the input flight plan track... and real-time flight track To be processed uniformly; This is the nth waypoint of the planned flight path. For the m-th track point of the real-time flight track, coordinate sampling and interpolation are performed on the flight plan track and the real-time flight track to make their number of tracks approximately equal; Each waypoint is formatted as a three-dimensional geographic location: , All waypoints have undergone coordinate transformation (WGS84) for spatial distance calculation. , , These are the longitude, latitude, and altitude of the i-th track point on the flight plan track, respectively. , , These are the longitude, latitude, and altitude of the j-th track point on the real-time flight track, respectively.

[0041] Distance metric function selection: for waypoint pairs The distance error between them is defined by the following three-dimensional spatial distance function:

[0042] in, This represents the spatial difference in the x-axis, y-axis, and z-axis directions after coordinate system transformation; The distance function can be switched to: great circle distance (ground projection) + height difference; pure 2D ground distance and 3D spatial Euclidean distance (default).

[0043] Dynamic Time Warping (DTW) Alignment Path Calculation: Constructing a Two-Dimensional Distance Matrix Each element is The minimum cumulative path cost is calculated using a dynamic programming algorithm.

[0044] in, To satisfy the alignment constraints of an effective path, The total cost of the path between the planned flight track and the real-time flight track; supports starting alignment window constraints to limit the maximum time alignment offset, improving efficiency and stability.

[0045] Track alignment score output: The total cost of the path With the set maximum acceptable alignment distance Normalization is performed to obtain the dynamic time-warped track morphology alignment score. :

[0046] The smaller the cost of track alignment, the closer the score is to 1, indicating that the overall shape of the track is closer; if the cost exceeds the set tolerance, the score will approach 0. (5) Overall consistency score: The scores from the above dimensions are weighted and fused to obtain a comprehensive consistency score, which serves as the basis for alarm determination; the weights for each scoring dimension are designed. ,satisfy Calculate the overall consistency score:

[0047] The higher the value, the better the consistency of the flight path.

[0048] This application embodiment also includes a scoring function configuration mechanism; the scoring function configuration mechanism includes: To enhance the system's adaptability under different operating scenarios, airspace conditions, and airline strategies, this application provides a flexible scoring function configuration mechanism for dynamically adjusting key parameters such as the weight, window length, and threshold range of various trajectory scoring indicators, and supports runtime hot updates without restarting the main system.

[0049] (1) Configuration item support and data structure; The scoring function consists of multiple sub-scoring items, and the system supports the definition and updating of the following parameters: Weighted values ​​for each sub-scoring item (horizontal, vertical, heading, DTW); scoring interval boundaries used for alarm level determination; maximum tolerance limit for DTW distance; sliding window time range (seconds); altitude error tolerance threshold; heading error tolerance threshold; (2) Loading mechanism and structural design: During startup, the scoring module loads the configuration file from the specified path and establishes a memory mapping. All subsequent scoring calculations read the memory parameters generated from the configuration file, eliminating the need for hard-coding. The module uses a configuration manager that periodically monitors changes to the configuration file and reloads the parameters when changes occur; it also provides a read-only interface for the scoring module to use. (3) Explanation of the mechanism supporting hot updates: To enable parameter adjustments without downtime, this module implements a hot update mechanism: a file monitoring mechanism; asynchronous loading is triggered upon detecting changes in file content; automatic rollback to the previous version is performed when loading fails to prevent invalid parameters from affecting operation; and configuration uploads via a web interface are supported (optional extension).

[0050] The implementation process of step 2 above includes: The alarm level determination module is used to determine the degree of deviation between the flight track and the flight plan based on the comprehensive consistency score, and outputs structured alarm information. This module can flexibly adopt two working modes. The alarm level determination module process is as follows: Figure 3 As shown.

[0051] (1) Supervision mode: The machine learning decision tree model is used to perform multi-dimensional fusion judgment on the overall consistency score, replacing the traditional threshold hard coding rule, and improving the system's adaptability and generalization ability to complex track deviation situations.

[0052] (2) Unsupervised mode (when decision tree is unavailable): When the model is unavailable or training data is lacking, the model will automatically be downgraded and a lightweight fine-tuning mechanism for the scoring threshold will be used to determine the level.

[0053] Decision tree fusion scoring mechanism: (1) Input feature design: The alarm level determination module takes as input the results of each scoring item output by the "track consistency scoring module" in the previous stage. Supported input is the comprehensive consistency score result, which can be derived from sources including but not limited to: horizontal position deviation score, vertical altitude deviation score, heading angle difference score, DTW track shape consistency score, and the average speed of the current track segment.

[0054] (2) Modeling of decision tree classifiers: The alarm level determination module uses a decision tree classifier trained with supervised learning to predict the alarm level based on the input comprehensive consistency score. Alarm levels are divided into four levels (or three / five levels are configurable): Level 0: Good consistency, no alarm required; Level 1: Slight deviation, attention recommended; Level 2: Moderate deviation, track correction recommended; Level 3: Severe deviation, immediate alarm and action required. The training data comes from historical flight track ratings and manually labeled levels, and can be customized according to specific operating areas or airline standards.

[0055] (3) Training and deployment mechanism: Training phase: The training set is built offline, with the input being the comprehensive consistency score; the labels are the consistency levels of historical human assessments; a classification tree model is used for fitting; and the tree model structure is output.

[0056] Deployment phase: After the model is loaded, the scoring results of each sliding window are judged in real time; the system supports calling the interface to complete the classification; the judgment result and the score value are output together for use by the subsequent alarm output optimization module; the system supports model version switching, canary deployment and other mechanisms.

[0057] (4) Decision transparency and human-machine explanation interface: The system supports returning the current track score input, the hit tree branch path, and the level output as structured JSON; it supports generating visual alarm description text by combining rules, for example: the current track overall deviation score is 0.58, the main reason is that the heading score is too low (0.30), alarm level: 2 (moderate deviation); the decision tree structure can be exported as a graph, which can be embedded in the debugging interface or used as system configuration review material.

[0058] Lightweight fine-tuning mechanism for scoring thresholds: To improve the adaptability of the flight consistency alarm system in scenarios such as lack of high-quality historical samples, unavailable supervised learning models, or unstable boundary judgments, this application designs a lightweight fine-tuning mechanism for the scoring threshold to dynamically adjust the mapping boundary between consistency scores and alarm levels, thereby optimizing the stability and reliability of alarm level output.

[0059] This mechanism is independent of decision trees and other models, and is suitable for some typical application scenarios: when there are no available models or labeled historical data in the early stage of system deployment; when sudden changes in the aircraft operating environment (such as crossing airspace or flight segments) cause drift in scoring features. The alarm level frequently changes near the boundary, affecting alarm reliability; the operation and maintenance requirements require parameters to be automatically adjusted without restarting.

[0060] (1) Input data: This mechanism accepts the following inputs: Real-time consistency score value: score_t; The score sequence S within the sliding time window is S=[score_{t n},…,score_{t 1}], where n is the nth time window; The initial alarm level threshold set of the system is T={τ_1,τ_2,…,τ_k}; Lightweight fine-tuning parameters (including floating threshold factor, jump suppression window length, etc.) are provided through configuration files; (2) Scoring trend analysis and boundary stability assessment: Within each alarm cycle, the system calculates the statistical characteristics of the scoring sliding window, including the mean μ, standard deviation σ, and alarm level jump frequency. If the score is found to be near a certain alarm level boundary for a long period of time, or if the level fluctuates frequently, it is determined that the current threshold setting cannot effectively distinguish the actual level status.

[0061] For example, if the initial threshold for level 2 is set to τ_2=0.7, but the score falls continuously between [0.68,0.72], causing level 1 and level 2 to change alternately, the threshold fine-tuning mechanism is triggered.

[0062] (3) Threshold fine-tuning strategy: The system makes minor adjustments to the alarm level thresholds based on the following logic:

[0063] in, The alarm level threshold before adjustment The adjusted alarm level threshold is represented by α, which is a fine-tuning scaling factor (e.g., 0.2). The standard deviation of the scoring sequence; This indicates a slight upward or downward adjustment; the direction of the adjustment is determined by the direction of the score's offset at the boundary.

[0064] The threshold adjustment has a constrained range to prevent excessive deviation. The system also has a recovery mechanism; if the score stabilizes, the threshold will slowly return to its original setting.

[0065] (4) Output results: This mechanism outputs a set of dynamically adjusted alarm level thresholds. This information is used by the subsequent alarm output optimization module. It can also include diagnostic information (such as fine-tuning direction, oscillation intensity, trigger flags, etc.) for evaluation by the log system.

[0066] Compared with static scoring methods, this mechanism has the following advantages: it improves the dynamic adaptability of the score to the grade determination boundary; it reduces the frequent jumps in grade output at the boundary, thus improving system stability; it is compatible with actual operating scenarios where pre-trained models cannot be used, such as cold start and spatial domain switching; and it supports hot updates of configuration parameter files, allowing for optimization without interrupting services.

[0067] Following step 2 above, an alarm output optimization module is also included. This module employs a multi-layered alarm output optimization mechanism to improve the accuracy of alarm output and user experience. The alarm output optimization mechanism includes: In flight consistency alarm systems, the lack of a sophisticated alarm post-processing mechanism can easily lead to problems such as frequent repeated alarms, false alarms due to short-term fluctuations, and multiple alarms for the same deviation behavior, affecting air traffic controllers' judgment and system stability. To address this, this application designs a multi-level alarm output optimization mechanism, including strategies such as alarm deduplication, time aggregation, level suppression, and sliding smoothing, to improve the accuracy of alarm output and user experience.

[0068] (1) Alarm deduplication mechanism: For the same flight target, the system generates a consistency score and level judgment in each evaluation window. If multiple consecutive windows generate alarms of the same level, the system outputs the alarm information only once to avoid duplicate alarm pushes. Deduplication judgment conditions include: unique aircraft identifier; alarm levels are consistent within adjacent time windows; and the time interval is less than the set deduplication threshold (e.g., 15 seconds).

[0069] (2) Time aggregation mechanism: The system introduces a time-based aggregation window to combine multiple alarm records within a short period into a single high-priority alarm. The aggregation logic is as follows: Set the aggregation window (e.g., 30 seconds); within the window, count the number of alarms, the maximum level, the average score, and other metrics; output the aggregated alarm, including start time, duration, maximum level, and average deviation score. (3) Alarm level suppression strategy: The system is designed with a delayed alarm level determination mechanism to avoid frequent changes in alarm levels due to slight fluctuations in scores. The determination mechanism is as follows: if the alarm level decreases in the current window compared to the previous window, the alarm will only be officially reduced after two consecutive windows of decrease; if the alarm level increases, a new alarm will be output immediately (prioritizing safety); the lag suppression time for alarm level changes (e.g., 2 windows) can be set in the configuration, with a default smoothing interval of approximately 10 seconds.

[0070] (4) Sliding smoothing mechanism (auxiliary): A moving average is applied to the scoring trend to further suppress scoring fluctuations; the smoothed score serves as a reference feature for determining the alarm level (with optional weights); the smoothing intensity can be adjusted according to the flight phase, for example, the response can be appropriately enhanced during the approach phase.

[0071] (5) Configurable and visual output: All optimization strategy parameters support YAML / JSON configuration; the optimization module output structure is unified, with fields including: alarm ID, aircraft ID, timestamp; level, score, alarm trigger rule, alarm type description; it can be integrated into ground terminals, automatic voice broadcasting systems, or console display interfaces.

[0072] This application also provides a flight consistency alarm system based on multi-dimensional score fusion and intelligent grading mechanism, implementing the flight consistency alarm method based on multi-dimensional score fusion and intelligent grading mechanism described in the above embodiments, which includes: The trajectory consistency scoring module is used to obtain a comprehensive consistency score based on multi-dimensional scoring to evaluate the consistency between the actual flight trajectory and the planned flight trajectory of the aircraft. The multi-dimensional scoring results include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-curved trajectory alignment score. The alarm level determination module is used to determine the degree of deviation between the actual flight track and the flight plan track based on the comprehensive consistency score, and output alarm information; the alarm information includes the alarm level.

[0073] Optionally, embodiments of this application also include an alarm output optimization module, which integrates the alarm output optimization mechanism described in the above embodiments.

[0074] This application's multi-dimensional scoring structure employs horizontal distance, vertical altitude, heading deviation, and DTW (Depth-to-Weight) morphological scoring, and all scoring functions support configurability. Alternative methods may include: Replace track similarity scores with other path comparison methods such as Hausdorff distance and Frechet distance; replace heading angle scores with attitude change rate or lateral offset integrals; replace vertical deviation scores with climb / descent rate integrals; and replace DTW scores with fixed template matching. This application employs a score fusion and decision tree model for grade output and supports intelligent degradation strategies. Alternative methods include: Replace decision trees with other classification algorithms such as random forests and support vector machines; use rule engines (such as Drools) to determine scoring thresholds; use deep neural networks or logistic regression for classification; and output alarm levels directly based on hard thresholds of scoring values.

[0075] This application proposes a flight consistency alarm method and system based on a "multi-dimensional track consistency scoring function + intelligent hierarchical mechanism," which possesses high interpretability, scalability, and engineering adaptability, and can effectively improve the accuracy and early warning of flight deviation detection. Its core technical points are as follows, constituting the main content to be protected in this application: (1) Multi-dimensional track consistency scoring mechanism: This application innovatively incorporates four dimensions of flight path (horizontal offset, vertical altitude, heading angle, and overall path shape) into a consistency scoring system. A normalized function maps each indicator to a range of 0–1, and a fusion scoring method is designed for overall consistency measurement. This mechanism introduces "shape alignment" capabilities on top of the original "point deviation" method, resulting in stronger anomaly detection capabilities.

[0076] (2) DTW track alignment score as a consistency feature: This application introduces the DTW (Dynamic Time Warping) algorithm into flight consistency scoring for the first time, capturing the trend difference between the nonlinear deviation of the flight plan and the actual trajectory in the time dimension, and providing a quantifiable indicator for overall morphology analysis.

[0077] (3) Configuration mechanism and hot update capability of scoring function: All parameters used in the scoring calculation (weighting coefficients, threshold ranges, sliding windows, etc.) can be configured externally via YAML or JSON files. The system can perform hot loading of parameters during operation, making it easy to quickly adapt to different airspaces, airports, or operating procedures.

[0078] (4) Hierarchical alarm judgment model based on score fusion results: This application proposes an alarm level judgment model based on scoring feature fusion, which supports hierarchical classification of fused features through a trained decision tree and outputs alarm levels of 0-3 (normal, minor, moderate, and severe), making it convenient for air traffic control systems to respond flexibly according to the level.

[0079] (5) Lightweight fine-tuning mechanism for scoring thresholds: This application designs a self-adjusting mechanism for scoring thresholds that does not require supervised data. Based on historical scoring trends, oscillation frequency, and level switching stability, it can automatically fine-tune and restore the level boundary values, ensuring robustness of judgment even in model-less or cold-start scenarios.

[0080] (6) Intelligent degradation judgment and switching strategy based on operating status: This application proposes an intelligent alarm judgment strategy switching mechanism, which differs from the traditional "single-path" judgment process. When the model is available, this mechanism prioritizes the use of a decision tree model for high-precision alarm level output, which has strong interpretability and classification performance. When the model is detected to be unavailable (e.g., not loaded, version invalid, inference abnormal) or the scoring features are abnormal (e.g., distribution drift, feature missing), the system will automatically downgrade to the scoring threshold judgment path to ensure stable and uninterrupted system operation.

[0081] (7) Alarm output optimization mechanism: To prevent high-frequency redundant alarms, this application proposes an alarm deduplication and aggregation algorithm based on time windows and level changes, which realizes continuity judgment, alarm buffering and debouncing processing, and optimizes the alarm display and push experience.

[0082] Meanwhile, this mechanism supports the perception of scoring trends and oscillation patterns, and dynamically adjusts the level boundaries in conjunction with the fine-tuning module during the downgrade process, realizing a smooth switch from "model-driven" to "rule-driven" and ensuring that the system has strong fault tolerance, interpretability and real-time performance.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism, characterized in that, include: Step 1: Based on the multi-dimensional scores, obtain a comprehensive consistency score to evaluate the consistency between the actual flight path and the planned flight path of the aircraft; the multi-dimensional scores include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped path alignment score. Step 2: Based on the comprehensive consistency score, determine the degree of deviation between the actual flight track and the planned flight track, and output alarm information; the alarm information includes the alarm level; Step 1 includes: Step 11: Calculate the horizontal distance score: Synchronize and match the actual flight track with the planned flight track to ensure a one-to-one correspondence between corresponding time points; calculate the geographical distance error under the horizontal projection; normalize all distance differences to obtain the horizontal distance score. Where 1 represents complete consistency and 0 represents maximum deviation; Step 12: Calculate the vertical altitude score: Extract actual flight altitude and planned altitude data and perform time synchronization matching; calculate the altitude difference between each pair of matching points; normalize the data according to the set maximum allowable altitude deviation to obtain the vertical altitude score. ; Step 13: Calculate the heading angle score: Extract the actual heading angle and the planned heading angle data and perform time synchronization matching; calculate the heading angle difference between each pair of matching points, with a value range of [0°, 180°]; normalize the angle difference to obtain the heading angle score. ; Step 14: Introduce the Dynamic Time Warping (DTW) algorithm to perform temporal spatial similarity alignment calculation between the flight plan track and the real-time flight track, and obtain the dynamic time warping track morphology alignment score; Step 15: Based on the horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped track alignment score obtained in Steps 11 to 14, a comprehensive consistency score is obtained; Step 2 includes: Step 21: The alarm level determination module receives the comprehensive consistency score sent by the track consistency scoring module and obtains the average speed of the current track segment; Step 22: Determine whether to use the decision tree module. If so, load the decision tree module, perform model inference, and output the alarm level. Otherwise, use a lightweight fine-tuning mechanism for the scoring threshold to perform scoring trend analysis and boundary stability judgment to obtain the alarm level threshold. Compare the overall consistency score with the alarm level threshold and output the alarm level.

2. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 1, characterized in that, Step 14 includes: Step 141: The system first processes the input flight plan track. and real-time flight track To be processed uniformly; This is the nth waypoint of the planned flight path. For the m-th trackpoint of the real-time flight track, coordinate sampling and interpolation are performed on the flight plan track and the real-time flight track to make the number of trackpoints approximately equal; the format of each trackpoint is unified to a three-dimensional geographic location. , All waypoints have undergone coordinate transformation for spatial distance calculation. , , These are the longitude, latitude, and altitude of the i-th track point on the flight plan track, respectively. , , These are the longitude, latitude, and altitude of the j-th trackpoint on the real-time flight path, respectively. Step 142: Define the following three-dimensional spatial distance function to represent track point pairs. Distance error between: ,in, This represents the spatial difference in the x-axis, y-axis, and z-axis directions after coordinate system transformation; Step 143: Construct a two-dimensional distance matrix Each of its elements is The minimum cumulative path cost is calculated using a dynamic programming algorithm. in, To satisfy the alignment constraints of an effective path, The total cost of the path between the planned flight track and the real-time flight track; Step 144: Calculate the total cost of the path using the following formula. With the set maximum acceptable alignment distance Normalization is performed to obtain the dynamic time-warped track morphology alignment score. : The smaller the cost of track alignment, the closer the score is to 1, indicating that the overall shape of the tracks is more similar; if the cost exceeds the set tolerance, the score will approach 0.

3. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 1, characterized in that, Step 15 includes: The scores from each dimension are weighted and fused together to obtain a comprehensive consistency score using the following formula, which serves as the basis for alarm determination: in, To achieve a comprehensive consistency score, the weights for the horizontal distance score, vertical altitude score, heading angle score, and dynamic time-warped track alignment score are as follows: And satisfy ; The higher the value, the better the consistency of the flight path.

4. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 1, characterized in that, It also includes a scoring function configuration mechanism; the scoring function configuration mechanism is used to dynamically adjust key parameters and supports runtime hot updates without restarting the main system; key parameters include the weights, window lengths, and threshold ranges of various track scores; The scoring function consists of multiple sub-scoring items, and the system supports the definition and updating of the following parameters: Weights of horizontal distance score, vertical altitude score, heading angle score, and dynamic time-normalized track alignment score; scoring interval boundaries used for alarm level determination; maximum tolerance limit for DTW distance; sliding window time range; altitude error tolerance threshold; heading error tolerance threshold; During startup, the trajectory consistency scoring module loads the configuration file in the specified path and establishes a memory mapping. All subsequent scoring calculations read the memory parameters formed by the configuration file, eliminating the need for hard coding. The trajectory consistency scoring module uses a configuration manager, which is responsible for periodically monitoring changes to the configuration file and reloading the parameters when changes occur. It provides a read-only interface for the trajectory consistency scoring module to call. The track consistency scoring module implements a hot update mechanism: a file monitoring mechanism; asynchronous loading is triggered after a change in file content is detected; automatic rollback to the previous version is performed when loading fails; and configuration can be uploaded via a web interface.

5. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 1, characterized in that, In step 22, the decision tree module is loaded to perform model inference and output alarm levels, including: Training phase: The training set is built offline, with the comprehensive consistency score as the input and the consistency level of historical manual assessment as the label. The CART classification tree model is used for fitting, and the tree model structure is output. Deployment phase: After the model is loaded, the score results of each sliding window are judged in real time. The system supports calling the interface to complete the classification. The judgment result and the score value are output together for use by the subsequent alarm output optimization module. The system supports returning the current track score input, the hit tree branch path, and the level output as structured JSON. In step 22, the input data for the lightweight fine-tuning mechanism of the scoring threshold is: the real-time consistency score value score_t; the scoring sequence S=[score_{t} within the sliding time window. n},…,score_{t 1}], n is the nth time window; the initial alarm level threshold set T={τ_1,τ_2,…,τ_k} is set by the system; lightweight fine-tuning parameters are provided through the configuration file; lightweight fine-tuning parameters include floating threshold factor and jump suppression window length; Scoring trend analysis and boundary stability judgment: In each alarm cycle, the system calculates the statistical characteristics of the scoring sliding window, including the mean μ, standard deviation σ, and alarm level jump frequency; if it is found that the score is still near the boundary of a certain alarm level after a preset time, or the level fluctuates frequently, it is determined that the current threshold setting cannot better distinguish the actual level status. If the current threshold setting cannot better distinguish the actual level status, the output of the scoring trend analysis and boundary stability judgment will be fine-tuned. The lightweight fine-tuning mechanism for the scoring threshold outputs the adjusted alarm level threshold. This information is used by the subsequent alarm output optimization module and can output diagnostic information for the log system to evaluate. The diagnostic information includes fine-tuning direction, oscillation intensity, and trigger flags. This is the threshold for the kth alarm level.

6. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 5, characterized in that, The threshold is fine-tuned using the following formula: in, The alarm level threshold before adjustment The adjusted alarm level threshold. To fine-tune the scaling factor, The standard deviation of the scoring sequence; This indicates a slight upward or downward adjustment; the direction of the adjustment is determined by the direction of the score's offset at the boundary; if the score tends to stabilize, the threshold will return to the original set value.

7. The flight consistency alarm method based on multi-dimensional scoring fusion and intelligent grading mechanism according to claim 1, characterized in that, Following step 2, step 3 is also included; step 3 includes: the alarm output optimization module improves the accuracy of alarm output and user experience through a multi-level alarm output optimization mechanism; the multi-level alarm output optimization mechanism includes alarm deduplication, time aggregation mechanism, alarm level suppression strategy and sliding smoothing mechanism; Alarm deduplication: For the same flight target, the system generates a consistency score and level judgment in each evaluation window; if multiple consecutive windows generate alarms of the same level, the system will only output the alarm information once to avoid repeated push of the same alarm; the deduplication judgment conditions include the unique identifier of the aircraft, the alarm level is consistent in adjacent time windows, and the time interval is less than the set deduplication threshold. Time aggregation mechanism: The system introduces a time aggregation window to aggregate multiple alarm records within a short period of time into a single high-priority alarm; the aggregation logic is as follows: set an aggregation window, count relevant indicators within the window, and output aggregated alarms; relevant indicators include alarm count, maximum level, and average score; aggregated alarms include start time, duration, maximum level, and average deviation score; Alarm level suppression strategy: The system is designed with an alarm level lag judgment mechanism. The judgment mechanism is as follows: if the alarm level decreases in the current window compared to the previous window, the alarm will only be officially reduced after two consecutive windows of decrease; if the level increases, a new alarm will be output immediately. Sliding smoothing mechanism: The scoring trend is averaged to suppress scoring fluctuations; the smoothed score is used as a reference feature to determine the alarm level, and thus weights are assigned. The output of the alarm output optimization module includes alarm ID, aircraft ID, timestamp, as well as level, score, alarm triggering rule, and alarm type description.

8. A flight consistency alarm system based on multi-dimensional score fusion and intelligent grading mechanism, implementing the flight consistency alarm method based on multi-dimensional score fusion and intelligent grading mechanism as described in any one of claims 1-7, characterized in that, include: The trajectory consistency scoring module is used to obtain a comprehensive consistency score based on multi-dimensional scoring to evaluate the consistency between the actual flight trajectory and the planned flight trajectory of the aircraft. The multi-dimensional scoring results include horizontal distance score, vertical altitude score, heading angle score, and dynamic time-curved trajectory alignment score. The alarm level determination module is used to determine the degree of deviation between the actual flight track and the flight plan track based on the comprehensive consistency score, and output alarm information; the alarm information includes the alarm level.

Images