Air route dynamic optimization method based on air traffic control data

Abstract

The application relates to the technical field of air traffic management, and discloses a flight route dynamic optimization method based on air traffic control data. The method first acquires real-time air traffic control data, including current aircraft state, airspace environment parameters and dynamic airspace event information; determines initial flight route parameters according to the first two parameters, performs hierarchical feature extraction on the dynamic airspace event information based on a multi-level airspace topology network, generates multi-level airspace event features; generates updated flight situation parameters through iterative feature fusion, determines a similar flight situation parameter set in combination with the similarity of historical flight situation parameters; generates a dynamic optimization prompt text in light of this, inputs a trajectory prediction model to obtain predicted flight route parameters, and finally determines a target optimized flight route in combination with preset constraint conditions. The method can respond to dynamic airspace changes in real time, and improves the adaptability and reliability of flight route optimization.

Description

Technical Field

[0001] This invention relates to the field of air traffic management technology, specifically a method for dynamic optimization of flight routes based on air traffic control data. Background Technology

[0002] With the rapid development of the air transport industry, air traffic volume continues to grow, and the airspace environment is becoming increasingly complex, placing higher demands on the optimization of flight routes. Traditional flight route planning methods are mostly based on static airspace information, which makes it difficult to respond to dynamically changing airspace environments in real time, leading to problems such as reduced flight efficiency and increased potential conflict risks.

[0003] In actual operation, the airspace environment is affected by various factors, such as weather changes, temporary air traffic control, and unforeseen aircraft situations. These dynamic airspace events alter the availability of airspace in real time. Existing optimization methods often lack effective hierarchical analysis mechanisms when dealing with these dynamic events, failing to comprehensively extract the event's characteristic information and resulting in inaccurate judgments of airspace status. Furthermore, when combining historical flight data for route optimization, the lack of accurate identification and utilization of similar flight situations makes it difficult to draw on past experience to generate reasonable optimization schemes, leading to insufficient adaptability and reliability of the optimization results.

[0004] Traditional trajectory prediction models rely heavily on labeled data during training, failing to adequately utilize the vast amount of unlabeled flight data and thus limiting their generalization ability. When faced with complex and ever-changing airspace scenarios, these models struggle to generate predicted routes that conform to actual constraints, further impacting the effectiveness of flight route optimization. These problems render existing flight route optimization methods inadequate for meeting the demands of modern air traffic management for efficiency, safety, and dynamic adaptability, necessitating a new technological solution. Summary of the Invention

[0005] The purpose of this invention is to provide a method for dynamic optimization of flight routes based on air traffic control data, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for dynamic optimization of flight routes based on air traffic control data, the method comprising:

[0007] Acquire real-time air traffic control data, which includes current aircraft status parameters, airspace environment parameters, and dynamic airspace event information;

[0008] Determine the initial flight path parameters based on the current aircraft status parameters and the airspace environment parameters;

[0009] Based on a multi-level airspace topology network, hierarchical feature extraction is performed on the dynamic airspace event information to generate multi-level airspace event features.

[0010] Iterative feature fusion processing is performed based on the initial flight path parameters and the multi-level airspace event characteristics to generate updated flight situation parameters.

[0011] Based on the similarity between the updated flight status parameters and the historical flight status parameters corresponding to other aircraft, a set of similar flight status parameters is determined.

[0012] Dynamic optimization prompt text is generated based on the updated flight situation parameters and the set of similar flight situation parameters;

[0013] The dynamic optimization prompt text is input into the trajectory prediction model, and the predicted flight path parameters are output.

[0014] The target optimized flight path is determined based on the predicted flight path parameters and the preset flight path constraints.

[0015] Preferably, the step of extracting hierarchical features from the dynamic airspace event information based on a multi-level airspace topology network to generate multi-level airspace event features includes:

[0016] In the multi-level airspace topology network, identify the airspace event entity to be processed at the current level;

[0017] The flight parameter feature transformation is performed on the airspace event entity to be processed to generate the airspace event feature vector of the current level;

[0018] In the multi-level airspace topology network, determine the airspace event entity to be processed in the next level of the current level;

[0019] Based on the airspace event entity to be processed at the next level and the airspace event feature vector at the current level, feature fusion processing is performed to generate target updated airspace event features;

[0020] The target updated airspace event features are used as updated airspace event features, the next level is used as the current level, and the step of determining the next level airspace event entity to be processed is returned to the execution until the feature fusion processing of the highest level is completed.

[0021] The update airspace event features corresponding to the highest level are used as the airspace event features of the multi-level system.

[0022] Preferably, the step of performing flight parameter feature transformation on the airspace event entity to be processed to generate an airspace event feature vector at the current level includes:

[0023] Obtain the spatial event description text corresponding to the spatial event entity to be processed;

[0024] The spatial event description text is processed by a pre-set spatial semantic encoder to output an event semantic feature vector;

[0025] The event semantic feature vector is transformed by flight parameter dimension to generate the airspace event feature vector of the current level.

[0026] Preferably, the step of performing feature fusion processing based on the next-level unprocessed airspace event entity and the current-level airspace event feature vector to generate target updated airspace event features includes:

[0027] The mean value of the spatial event feature vectors corresponding to all spatial event entities to be processed in the next level is calculated to generate secondary spatial event features.

[0028] The secondary airspace event features and the updated airspace event features are weighted and fused to generate the target updated airspace event features.

[0029] Preferably, determining the set of similar flight status parameters based on the similarity between the updated flight status parameters and the historical flight status parameters corresponding to other aircraft includes:

[0030] Calculate the airspace state similarity between the updated flight situation parameters and each historical flight situation parameter;

[0031] Filter specific historical flight status parameters whose airspace state similarity exceeds a preset similarity threshold;

[0032] The historical flight path parameters corresponding to the specific historical flight situation parameters are added to the set of similar flight situation parameters.

[0033] Preferably, the step of generating dynamic optimization prompt text based on the updated flight situation parameters and the set of similar flight situation parameters includes:

[0034] The current flight constraints are determined based on the updated flight status parameters.

[0035] Extract similar flight path trajectory data from the set of similar flight situation parameters;

[0036] The current flight constraints and the similar flight path trajectory data are combined to synthesize path description text, generating the dynamic optimization prompt text.

[0037] Preferably, before acquiring real-time air traffic control data, the method further includes:

[0038] Constructing a multi-level airspace topology network includes:

[0039] Acquire historical airspace operation data, which includes aircraft trajectory instances, airspace control event instances, and flight conflict instances;

[0040] Determine the airspace topology relationships between the aircraft trajectory instance, the airspace control event instance, and the flight conflict instance;

[0041] The multi-level spatial topology network is generated based on the spatial topology relationship.

[0042] Preferably, the trajectory prediction model is obtained through the following steps:

[0043] Obtain a flight training sample set, which includes labeled flight route parameters;

[0044] The trajectory prediction model is used to process the sample dynamic optimization prompt text and output the predicted flight path parameters.

[0045] The parameters of the trajectory prediction model are adjusted based on the difference between the predicted flight path parameters and the labeled flight path parameters.

[0046] Preferably, before processing the sample dynamic optimization prompt text through the trajectory prediction model, the method further includes:

[0047] Obtain flight pre-training samples without labeled flight status parameters;

[0048] The flight pre-training samples are subjected to comparative learning feature encoding to generate flight comparative feature vectors;

[0049] The flight comparison feature vector is processed by the initial trajectory prediction model to output predicted flight feature parameters;

[0050] The model pre-training loss value is determined based on the predicted flight feature parameters and the flight comparison feature vector;

[0051] The parameters of the initial trajectory prediction model are optimized based on the model pre-training loss value to obtain a pre-trained trajectory prediction model.

[0052] Preferably, the step of performing contrastive learning feature encoding on the flight pre-training samples to generate flight contrastive feature vectors includes:

[0053] Identify airspace topology patterns in the flight pre-training samples;

[0054] Match the core flight feature encoding corresponding to the airspace topology pattern;

[0055] Generate a comparative feature code for the core flight feature code;

[0056] The core flight feature encoding and the comparison feature encoding are coupled to generate the flight comparison feature vector.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] By acquiring real-time air traffic control data, which includes current aircraft status parameters, airspace environment parameters, and dynamic airspace event information, comprehensive and real-time basic information is provided for the dynamic optimization of flight routes, enabling timely capture of various changes in the airspace.

[0059] When processing dynamic airspace event information, a multi-level airspace topology network is used for hierarchical feature extraction to generate multi-level airspace event features. This hierarchical processing approach can deeply mine the features of dynamic airspace events from different levels, taking into account both the details of local airspace events and the correlations between events at different levels. This results in a more comprehensive and in-depth understanding of dynamic airspace events, providing a more accurate basis for subsequent route optimization.

[0060] By iteratively fusing initial flight path parameters and multi-level airspace event features, updated flight situation parameters are generated, achieving an effective combination of initial path and dynamic event features. This enables dynamic adjustment of the judgment of the flight situation, making the description of the flight situation more consistent with the real-time airspace conditions.

[0061] When determining the set of similar flight situation parameters, the similarity between the updated flight situation parameters and the historical flight situation parameters of other aircraft can be used to accurately select historical flight data that can be referenced. This similar historical data contains past flight experience in similar airspace situations, providing valuable reference for current route optimization and helping to generate more adaptive optimization schemes.

[0062] Dynamic optimization prompts are generated based on updated flight situation parameters and sets of similar flight situation parameters. These prompts are then input into a trajectory prediction model, which outputs predicted flight path parameters. Combined with preset constraints, the target optimized flight path is determined, effectively integrating historical experience with real-time data. This approach fully utilizes valuable information from historical flight data, while incorporating real-time dynamic adjustments, enabling the generated target optimized flight path to better adapt to complex and ever-changing airspace environments.

[0063] In the process of constructing a multi-level airspace topology network, the airspace topology relationships between various instances are determined based on historical airspace operation data, providing a reasonable network structure foundation for hierarchical feature extraction and ensuring the effectiveness and systematic nature of feature extraction.

[0064] During the training process of the trajectory prediction model, not only labeled flight training samples were utilized, but unlabeled flight pre-training samples were also processed through comparative learning. This fully leverages the information in a large amount of unlabeled data, improving the model's generalization ability. When facing different airspace scenarios, the model can more flexibly generate predicted routes that meet actual needs, enhancing the reliability of route optimization. Attached Figure Description

[0065] Figure 1 This is a time sequence diagram of the flight route dynamic optimization method based on air traffic control data described in this invention.

[0066] Figure 2 Flowchart for generating multi-level airspace event characteristics;

[0067] Figure 3 A flowchart for generating feature vectors for spatial domain events;

[0068] Figure 4 A flowchart for determining a set of similar flight situation parameters;

[0069] Figure 5 A flowchart for constructing a multi-level airspace topology network. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Please see Figure 1 The present invention provides a method for dynamic optimization of flight routes based on air traffic control data, the method comprising:

[0072] This invention discloses a method for dynamic optimization of flight routes based on air traffic control data, the overall implementation of which is as follows:

[0073] The system acquires real-time air traffic control data, including current aircraft status parameters such as real-time three-dimensional coordinates, instantaneous ground speed, heading angle, climb rate, and remaining fuel; airspace environmental parameters include environmental elements such as real-time wind speed vector, visibility level, cloud height, pressure field distribution, and terrain elevation data; and dynamic airspace event information includes timely information such as temporary restricted area designation notices, air traffic flow saturation warnings, weather change alerts, and aircraft abnormal status reports.

[0074] The initial flight route parameters are determined based on the current aircraft status parameters and airspace environment parameters. Combining the coordinate references of the aircraft's takeoff airport and destination airport, referencing the standard instrument departure procedure and standard instrument approach procedure, and incorporating the limitations of real-time weather conditions on the cruise altitude, an initial route plan is generated, including the latitude and longitude of the navigation points along the route, the expected magnetic heading for each segment, the planned altitude sequence, and the expected airspeed for each stage. The spacing between navigation points is dynamically adjusted according to the airspace type and aircraft performance characteristics.

[0075] Based on a multi-level airspace topology network, hierarchical feature extraction is performed on dynamic airspace event information to generate multi-level airspace event features. The multi-level airspace topology network is divided into five levels according to airspace management responsibilities: national control area, regional control area, terminal control area, approach control area, and tower control area. Each level includes basic attributes such as corresponding airspace unit boundaries, control sector divisions, and communication frequency coverage. The event features are transmitted and aggregated layer by layer through inter-level association mapping.

[0076] Iterative feature fusion processing is performed based on initial flight path parameters and multi-level airspace event characteristics to generate updated flight situation parameters. A gated recurrent unit is used to temporally encode the initial path parameter sequence, while a convolutional neural network is used to compress the spatial dimension of the multi-level event features. Attention mechanisms are used in the feature fusion layer to assign weights to different features. After three iterations, a comprehensive situation parameter set is formed, including spatial position deviation, time delay risk, and conflict probability index.

[0077] Based on the similarity between updated flight situation parameters and historical flight situation parameters corresponding to other aircraft, a set of similar flight situation parameters is determined. A six-dimensional feature space including position, altitude, speed, weather conditions, and airspace event type is constructed. The similarity metric between the current situation and historical situations is calculated using Mahalanobis distance. A dynamic threshold is set to filter historical records that meet the criteria, forming a similar case library.

[0078] Dynamic optimization prompts are generated based on updated flight situation parameters and sets of similar flight situation parameters. Key constraints in the current situation are extracted, including minimum safe altitude, maximum turning bank angle, and no-fly zone boundary coordinates. Simultaneously, flight path adjustment segments from similar cases are extracted, and structured data is transformed into a text sequence containing spatial descriptions, temporal instructions, and constraints using a natural language generation model.

[0079] The dynamically optimized prompt text is input into the trajectory prediction model, which outputs predicted flight path parameters. The trajectory prediction model adopts an encoder-decoder architecture. The encoder performs semantic parsing of the text prompt, and the decoder outputs a sequence of waypoints for the next 120 minutes. Each waypoint contains four parameters: timestamp, latitude and longitude, altitude, and ground speed.

[0080] The optimal flight path for the target is determined based on the predicted flight path parameters and preset flight path constraints. The preset constraints cover aircraft performance limitations, air traffic rules, and airspace usage restrictions. The predicted path is verified and corrected segment by segment using a constraint satisfaction verification algorithm, ultimately generating an executable optimized path.

[0081] Example 1: See Figure 2 Based on a multi-level airspace topology network, hierarchical feature extraction is performed on dynamic airspace event information. When generating multi-level airspace event features, the first step is to identify the airspace event entities to be processed at the current level within the multi-level airspace topology network. The multi-level airspace topology network is divided into levels according to the actual airspace management architecture. Each level covers a specific geographical area and management responsibilities, and the types of airspace events and processing priorities differ between levels. The identification of airspace event entities to be processed requires considering the specific location of the event, its impact range, and the airspace management scope to which it belongs. Through spatial coordinate matching and attribute label identification, the event is accurately mapped to the corresponding level. Each level maintains an event queue, sorted according to the event reception time and impact level to ensure orderly subsequent processing.

[0082] The process involves transforming the flight parameter features of the airspace event entity to be processed, generating an airspace event feature vector for the current level. This process requires extracting key attributes of the event, including its start time, expected duration, geographical boundaries of the affected area, restrictions on the types of aircraft involved, and associated weather condition changes. These attributes encompass multiple dimensions such as time, space, and physical limitations, requiring standardization to convert different types of data into a unified numerical form and eliminate the impact of dimensional differences. For example, the time attribute is converted into seconds relative to a certain reference time, the geographical boundary is converted into a sequence of latitude and longitude coordinates, and the degree of impact is converted into a corresponding numerical value according to a preset level standard. After transformation, these attributes are combined to form a fixed-length vector, with each element corresponding to a feature, fully reflecting the core information of the airspace event.

[0083] In a multi-level spatial topology network, the entities of spatial events to be processed in the next level are identified. There are explicit inclusion or association relationships between levels, and the spatial scope of a higher level typically covers multiple spatial units in the next lower level. Based on the influence range and spatial location of the event at the current level, all spatial units in the next lower level associated with the current event are retrieved through hierarchical mapping relationships. From these next lower level spatial units, spatial events that are occurring or about to occur are selected to form the set of events to be processed in the next lower level. This process needs to consider spatial overlap and event correlation to ensure that events in the next lower level have a real impact or association with events in the current level, and to avoid the introduction of irrelevant events.

[0084] Feature fusion processing is performed on the spatial event entities to be processed at the next level and the feature vectors of spatial events at the current level to generate target updated spatial event features. First, the same feature transformation is performed on each event to be processed at the next level to obtain its own feature vector. Then, based on the degree of correlation between these events and the events at the current level, corresponding weights are assigned to the feature vectors of each next-level event. The degree of correlation can be calculated comprehensively based on factors such as spatial distance, overlap of influence range, and similarity of event types. Next, these weighted feature vectors of the next-level events are aggregated to form a comprehensive secondary feature vector. This secondary feature vector is then fused with the feature vectors of the spatial events at the current level. The fusion process uses a weighted summation of vector elements, with weights set according to the importance of the level and the influence of the event, ultimately generating a new feature vector, i.e., the target updated spatial event features.

[0085] The process involves updating the target spatial event features, using the next level as the current level, and then returning to determine the next level of spatial event entities to be processed, continuing until the highest level's feature fusion is complete. Each time a level is advanced, the event retrieval, feature transformation, and fusion process is repeated. During this iteration, each level's features absorb relevant event information from the next level, continuously enriching and refining the feature vector, which is then passed upwards. As the level increases, the spatial range covered by the feature vector expands, and the granularity of event information adjusts accordingly, gradually aggregating from specific local events to more macroscopic regional event features. In each iteration, the fused feature vector also needs to be normalized to ensure its numerical range remains stable, preventing numerical overflow or feature distortion after multiple fusions.

[0086] The updated airspace event features corresponding to the highest level are used as multi-level airspace event features. When the iteration process reaches the preset highest level, the updated airspace event features have integrated all relevant airspace event information from the lowest to the highest level, encompassing event features of different granularities and ranges. This feature vector comprehensively reflects the overall situation of various events in the current airspace environment, including both subtle local event changes and large-scale event trends. After generating the final multi-level airspace event features, feature selection algorithms can be used to filter elements in the vector, retaining features that have a greater impact on subsequent processing and removing redundant or secondary features, further optimizing the quality of the feature vector and making it more suitable for dynamic optimization analysis of flight routes.

[0087] Example 2: See Figure 3When performing flight parameter feature transformation on the airspace event entity to be processed to generate the airspace event feature vector at the current level, the airspace event description text corresponding to the airspace event entity to be processed is first obtained. This text information usually comes from real-time notifications from the air traffic control system, early warning information from the meteorological department, and temporary notices from airspace management, covering the basic attributes of the event, such as the specific time of the event, the airspace range involved, the nature of the event (such as sudden weather changes, temporary no-fly zones, traffic control, etc.), the expected duration, and the type of impact that may be caused to flights. The text content must be organized according to a unified format specification to ensure the completeness and accuracy of key information and avoid ambiguous or vague expressions so that subsequent semantic analysis can accurately capture the core content of the event.

[0088] A pre-built airspace semantic encoder processes the text describing airspace events, outputting an event semantic feature vector. This encoder, built on a deep neural network architecture and trained on a large amount of aviation text data, is capable of understanding the specific meanings of aviation terminology and the contextual information of event descriptions. The processing first segments the text into several basic semantic units (such as words or sub-phrases), then uses an embedding layer to convert each semantic unit into a low-dimensional vector representation, capturing its basic semantic information. Next, a multi-layered encoding structure processes these vectors, considering the sequence and relationships between semantic units, such as the causal relationship between "thunderstorm area" and "no entry," and the logical relationship between "traffic saturation" and "speed limit instruction." Through this processing, the text information is converted into a fixed-dimensional event semantic feature vector, where each element corresponds to different dimensions of semantic information in the text, comprehensively reflecting the deeper meaning of the event description.

[0089] The event semantic feature vector undergoes flight parameter dimensionality transformation to generate the airspace event feature vector for the current level. Since the event semantic feature vector primarily reflects the semantic information of the text, its dimension and numerical range do not perfectly match the feature space of flight parameters (such as position, velocity, and altitude). Therefore, dimensionality transformation is necessary to enable the generated feature vector to be directly fused with aircraft state parameters and airspace environment parameters. The transformation process is implemented using a pre-defined mapping matrix, constructed based on the feature space characteristics of flight parameters and the distribution patterns of semantic features. This matrix adjusts the dimension of the semantic feature vector to a dimension compatible with the flight parameter feature space. During the transformation, it is crucial to preserve key information in the semantic features, such as the scope and degree of the event's impact, while ensuring that the transformed feature vector maintains consistency in numerical distribution with other flight parameters to facilitate subsequent feature fusion.

[0090] Based on the feature fusion processing of the unprocessed airspace event entities at the next level and the airspace event feature vectors at the current level, when generating target update airspace event features, the mean of the airspace event feature vectors corresponding to all unprocessed airspace event entities at the next level is first calculated to generate secondary airspace event features. When there are multiple unprocessed events at the next level, each event corresponds to an airspace event feature vector, and these vectors reflect the feature information of different events. During the mean calculation process, these vectors need to be averaged according to their corresponding dimensions, that is, the mean of each dimension is equal to the sum of all vector elements in that dimension divided by the number of vectors. For some special cases, such as when the impact range of an event is much larger than that of other events, or its urgency is significantly higher, a higher weight can be assigned to it during the mean calculation to highlight its influence in the secondary features. Through such mean calculation, the feature information of multiple events at the next level is aggregated into a secondary airspace event feature vector, which can reflect the overall situation of events at the next level.

[0091] The secondary airspace event features and the current-level airspace event feature vector are weighted and fused to generate target-updated airspace event features. During the fusion process, different weights need to be assigned to the two feature vectors based on the importance of the level and the impact of the event. Generally, the event features at the current level reflect a more macroscopic or direct impact and have relatively higher weights, while the secondary features at the next lower level serve as supplementary information and have relatively lower weights. The specific weight values ​​are determined based on the actual needs of airspace management and the analysis results of historical data, and can be dynamically adjusted according to actual conditions. The weighted fusion calculation formula is as follows:

[0092] in: This indicates that the target updates the airspace event characteristics. This represents the spatial event feature vector at the current level. Represents the feature vector of secondary spatial domain events. This represents the weight coefficient of the feature vector at the current level. This represents the weight coefficients of the secondary eigenvectors, and Through this weighted summation, the features of the current level are organically integrated with the aggregated features of the next level. The resulting target updated spatial event features retain the main information of the current level events while integrating the relevant impacts of the next level events, thus providing a more comprehensive reflection of the overall situation of spatial events. After fusion, the target updated spatial event features need to be standardized to adjust their numerical range to a preset interval, avoiding excessive numerical differences that could affect the accuracy of subsequent calculations.

[0093] Example 3: See Figure 4To determine the set of similar flight status parameters based on the similarity between updated flight status parameters and historical flight status parameters corresponding to other aircraft, the airspace state similarity between the updated flight status parameters and each historical flight status parameter is first calculated. Updated flight status parameters include the current aircraft's real-time position, altitude, cruise speed, heading angle, remaining fuel, as well as information such as weather conditions, air traffic density, and surrounding airspace restrictions. Historical flight status parameters are records of past aircraft's states in similar airspace environments, containing various attributes corresponding to the updated flight status parameters. When calculating similarity, each parameter needs to be compared one by one, converting position parameters into latitude and longitude coordinate differences, speed parameters into vector differences, weather parameters into grade difference values, and traffic flow parameters into density deviation rates, etc. Through these conversions, different types of parameters are unified into quantifiable difference indicators, and then the overall airspace state similarity is obtained through comprehensive calculation.

[0094] Specific historical flight status parameters with airspace similarity exceeding a preset similarity threshold are selected. The preset similarity threshold is dynamically set based on airspace complexity. In open high-altitude airspace, the threshold can be appropriately lowered to include more reference cases; in complex airspace such as airport terminal areas and air corridors, the threshold is correspondingly increased to ensure that the selected historical cases highly match the current situation. During the selection process, the similarity value of each historical flight status parameter must be judged, and only historical parameters exceeding the threshold are considered specific historical flight status parameters. Furthermore, to avoid overly simplistic selection results, it is necessary to ensure that the specific historical flight status parameters cover different flight phases and different combinations of meteorological conditions.

[0095] Historical flight path parameters corresponding to specific historical flight situation parameters are added to a set of similar flight situation parameters. Historical flight path parameters include complete flight track data of past aircraft under the corresponding situation, such as the coordinates of navigation points passed, flight time for each segment, altitude adjustment records, speed change curves, and the number of heading corrections. These parameters need to be structured according to time sequence and spatial coordinates, removing outlier data and duplicate records. When adding them to the set, each historical flight path parameter needs to be labeled with corresponding historical flight situation characteristics, such as the air traffic control instructions at the time and emergency weather response measures.

[0096] When generating dynamic optimization prompts based on updated flight situation parameters and similar flight situation parameter sets, the current flight constraints are first determined based on the updated flight situation parameters. Current flight constraints include hard constraints and flexible constraints. Hard constraints include the geographical boundaries of no-fly zones (polygonal areas defined by latitude and longitude coordinates), minimum safe altitudes (determined based on terrain elevation and obstacle heights), and maximum flight speed limits (set based on airspace traffic and aircraft performance). Flexible constraints include allowable deviations from recommended flight routes, avoidance suggestions for traffic-intensive areas, and the timing of crossing weather-affected areas. These constraints need to be dynamically generated based on real-time data from the updated flight situation parameters. For example, when a new thunderstorm area is detected, it must be immediately included in the no-fly zone within the hard constraints.

[0097] Extract similar flight path trajectory data from a set of similar flight situation parameters. The extraction process should focus on trajectory segments that match the current flight phase. For example, if the aircraft is currently in the cruise phase, the focus should be on extracting trajectory adjustment data from historical cases during the cruise phase. Trajectory data includes the coordinates of key decision points, environmental parameters at the time of decision, and the effect of trajectory changes after the decision. Simultaneously, the degree of deviation between these similar trajectories and the current initial flight path needs to be calculated, retaining trajectory data with deviations within a reasonable range.

[0098] The process involves synthesizing path description text from current flight constraints and similar flight trajectory data to generate dynamic optimization prompts. The synthesis process uses a structured language template, first clearly listing all current constraints in a clear format such as "Prohibited Area: XXX latitude and longitude to XXX latitude and longitude" and "Minimum Safe Altitude: XXX meters." Then, based on similar flight trajectory data, it extracts suggested route adjustments, such as "Turning at XXX degrees North latitude and XXX degrees East longitude can avoid the traffic congestion area 50 kilometers ahead; based on historical cases, this turn can reduce flight time by XXX minutes," and "It is recommended to maintain level flight at an altitude of XXX meters, and continue climbing after passing through the cloud cover, consistent with the optimal trajectory under similar historical weather conditions." The text synthesis must ensure concise and accurate language, conform to aviation communication standards, and avoid vague expressions. A comprehensive dynamic optimization prompt text is generated through weighted fusion, using the following formula:

[0099] in: This represents the comprehensive feature value of dynamically optimized suggestion text. The characteristic value representing the current flight constraints, Feature values ​​representing similar flight path trajectory data This represents the weighting coefficient of the current flight constraints. The weighting coefficients represent the weighting coefficients of similar flight path trajectory data, and This formula ensures that the prompt text strictly adheres to current constraints while fully incorporating historical experience, resulting in optimized guidelines that are both standardized and practical.

[0100] Example 4: See Figure 5 I will describe in detail the implementation of Example 4, using specific examples, from two aspects: constructing a multi-level airspace topology network and training a trajectory prediction model. This will include a related data table.

[0101] Before acquiring real-time air traffic control data, constructing a multi-level airspace topology network requires several steps. First, historical airspace operation data is acquired, including aircraft trajectory instances, airspace control event instances, and flight conflict instances. Aircraft trajectory instances can be detailed trajectories of all flights operated by an airline on the Beijing-Shanghai route over the past six months. The trajectory data for each flight is recorded chronologically, including latitude, longitude, altitude, and ground speed information per minute. Airspace control event instances may include temporary control notices issued during this period, such as restrictions on passage through a certain airspace for a specific time period due to military activities. These notices will specify the time frame of the control and the coordinates of the affected airspace. Flight conflict instances may include records of incidents where two aircraft came too close to each other near the route, including the time of the conflict, the positions and speeds of the two aircraft at the time.

[0102] Analysis reveals that some flight trajectories bypass controlled airspace within specific timeframes, establishing a spatial avoidance relationship between trajectories and control events. For aircraft trajectory instances and flight conflict instances, it's possible that an aircraft's trajectory, due to untimely speed adjustments, intersects with another aircraft's trajectory at a certain airspace point, leading to a conflict. This constitutes a causal link between the trajectory and the conflict event. Such analysis allows us to clarify the relationships between various types of instances.

[0103] A multi-level airspace topology network is generated based on these airspace topology relationships. Different levels of the network correspond to different airspace management areas. For example, the national airspace level has a wide coverage, including multiple regional airspace units; regional airspace includes several terminal control areas, etc. Nodes in each level represent corresponding airspace units or events, and the edges between nodes represent their relationships, such as inclusion relationships, temporal relationships, etc.

[0104] The training of the trajectory prediction model also needs to be carried out step by step. First, a flight training sample set is obtained, which contains labeled flight route parameters. These samples can be flight data of the Beijing-Shanghai route under different seasons and weather conditions. The labeled flight route parameters are determined by experienced air traffic controllers based on the actual situation, including the optimal waypoint, altitude layer, etc.

[0105] The model processes the dynamic optimization prompt text for the sample flight and outputs predicted flight path parameters. The dynamic optimization prompt text includes weather conditions during the sample flight, such as visibility of 10 km and wind speed of 5 m / s, as well as airspace restriction information, such as airspace control from 8:00 AM to 10:00 AM. After processing this text, the model generates the corresponding predicted flight path parameters.

[0106] The model parameters are adjusted based on the differences between the predicted and labeled flight path parameters. Through continuous adjustments, the model's predictions become more accurate.

[0107] Table 1: A simplified table of aircraft trajectory examples from some historical airspace operation data.

[0108]

[0109] When constructing a multi-level airspace topology network, the trajectory data in the table above, combined with airspace control events and flight conflict instances, is used to identify the relationships between various instances, thereby generating the network. This trajectory data then becomes part of the training samples when training the trajectory prediction model, helping the model better learn and predict flight path parameters. In this way, both the construction of the multi-level airspace topology network and the training of the trajectory prediction model can be based on actual flight data, adapting to complex airspace environments and flight requirements.

[0110] Example 5: Before processing the dynamic optimization prompt text of the samples using the trajectory prediction model, it is necessary to obtain flight pre-training samples without labeled flight status parameters. These samples cover flight data from different regions and time periods, including the flight trajectory of the aircraft under various weather conditions, detailed information on airspace structure, and corresponding environmental parameters. For example, they include flight paths of the aircraft in different weather conditions such as sunny, rainy, and foggy days; trajectory records over different terrains such as mountains, plains, and oceans; and flight data in busy and open airspace. These samples do not contain manually labeled flight status parameters, but only retain the original flight data information, such as the aircraft's position coordinates, flight altitude, flight speed, and heading angle recorded at regular intervals, along with meteorological data such as wind speed, wind direction, and visibility at the time, as well as the distribution of other aircraft in the airspace.

[0111] When generating flight contrastive feature vectors by performing contrastive learning feature encoding on pre-trained flight samples, the sample data is first preprocessed to remove outliers, such as position jumps or velocity anomalies caused by equipment malfunctions, ensuring sample reliability. Then, data augmentation techniques are used to diversify the samples. For example, the same flight trajectory may be slightly stretched or compressed in the time dimension to simulate subtle changes in flight speed; the spatial position of the trajectory may be slightly offset to simulate differences in flight paths of different aircraft in the same airspace; or small random perturbations may be added to the trajectory to simulate minor heading adjustments that may occur in actual flight. Through these processes, multiple similar but not identical variant samples are generated for each original sample.

[0112] The initial trajectory prediction model processes the flight comparison feature vector and outputs predicted flight feature parameters. The network structure of the initial trajectory prediction model contains multiple hidden layers. The input layer receives the flight comparison feature vector, which is then processed by the multi-layer neural network. The output layer generates the predicted flight feature parameters. These predicted parameters include information such as the aircraft's likely position, altitude, and speed over a future period. During processing, the model transforms and extracts the input features through the connection weights between neurons in each layer, gradually learning the latent patterns in the samples, such as the typical flight trajectory characteristics of aircraft under different weather conditions and the influence patterns of airspace structure on flight paths.

[0113] The model pre-training loss value is determined based on the predicted flight feature parameters and the flight contrast feature vector. The predicted flight feature parameters output by the model are compared with the actual flight data contained in the flight contrast feature vector, and the difference between the two is calculated. The calculation of the difference involves multiple dimensions, such as the deviation of position coordinates, the difference in altitude, and the difference in speed. A total loss value is obtained by combining the differences across these dimensions. This loss value reflects the degree of deviation between the model's prediction and the actual situation; the smaller the loss value, the better the model's prediction performance.

[0114] The parameters of the initial trajectory prediction model are optimized based on the pre-training loss value to obtain a pre-trained trajectory prediction model. During optimization, the connection weights and bias parameters of each layer of the model are adjusted according to the calculated loss value. The loss value is propagated back from the output layer to the input layer using the backpropagation algorithm, adjusting the parameters layer by layer to reduce the difference between the prediction results and the actual data. This process is repeated iteratively, with each iteration using a portion of the pre-training samples to train the model, calculate the loss value, and adjust the parameters until the loss value stabilizes at a low level. At this point, the model can better capture the flight patterns in the samples, forming the pre-trained trajectory prediction model.

[0115] When generating flight comparison feature vectors by performing comparative learning feature encoding on pre-trained flight samples, the first step is to identify airspace topology patterns in the samples. By analyzing a large amount of flight trajectory data, different airspace usage patterns are summarized, such as the flight traffic distribution characteristics of certain airspaces within a specific time period, common path patterns of aircraft crossing airspace boundaries, and differences in flight trajectories of different types of aircraft within the same airspace. These patterns are automatically classified using clustering algorithms, grouping flight samples with similar characteristics into the same pattern category.

[0116] Core flight feature codes are assigned to match airspace topology patterns. A unique core feature code is assigned to each identified airspace topology pattern. This code contains key feature information about the flight trajectory under that pattern, such as typical altitude variation ranges, common speed intervals, and major heading change nodes. The core feature code is obtained by statistically analyzing and refining the features of all samples under that pattern, and it represents the essential characteristics of that pattern.

[0117] A contrastive feature code is generated based on the core feature code. This contrastive feature code is obtained by controllingly adjusting certain features within the core code. For example, minor fluctuations are made within the altitude range reflected by the core code, subtle adjustments are made within typical speed ranges, or the positions of heading change nodes are slightly offset. These adjustments ensure that the contrastive code differs from the core code while maintaining a high degree of similarity, reflecting subtle changes within the same airspace topology pattern.

[0118] Feature coupling is performed on the core flight feature encoding and the contrast feature encoding to generate a flight contrast feature vector. The core encoding and the contrast encoding are then combined according to certain rules, such as concatenating their feature elements in sequence or calculating their correlation values ​​in different dimensions, to form a comprehensive feature vector. This process ensures that the feature vector contains both the core features of the airspace topology pattern and information on variations within the pattern, providing richer feature inputs for subsequent model pre-training and helping the model better learn flight patterns under different airspace topologies.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic optimization of flight routes based on air traffic control data, characterized in that, include: Acquire real-time air traffic control data, which includes current aircraft status parameters, airspace environment parameters, and dynamic airspace event information; Determine the initial flight path parameters based on the current aircraft status parameters and the airspace environment parameters; Based on a multi-level airspace topology network, hierarchical feature extraction is performed on the dynamic airspace event information to generate multi-level airspace event features. Iterative feature fusion processing is performed based on the initial flight path parameters and the multi-level airspace event characteristics to generate updated flight situation parameters. Based on the similarity between the updated flight status parameters and the historical flight status parameters corresponding to other aircraft, a set of similar flight status parameters is determined. Dynamic optimization prompt text is generated based on the updated flight situation parameters and the set of similar flight situation parameters; The dynamic optimization prompt text is input into the trajectory prediction model, and the predicted flight path parameters are output. The target optimized flight route is determined based on the predicted flight route parameters and the preset flight route constraints. The method based on a multi-level airspace topology network performs hierarchical feature extraction on the dynamic airspace event information to generate multi-level airspace event features, including: In the multi-level airspace topology network, identify the airspace event entity to be processed at the current level; The flight parameter feature transformation is performed on the airspace event entity to be processed to generate the airspace event feature vector of the current level; In the multi-level airspace topology network, determine the airspace event entity to be processed in the next level of the current level; Based on the airspace event entity to be processed at the next level and the airspace event feature vector at the current level, feature fusion processing is performed to generate target updated airspace event features; The target updated airspace event features are used as updated airspace event features, the next level is used as the current level, and the step of determining the next level airspace event entity to be processed is returned to the execution until the feature fusion processing of the highest level is completed. The update airspace event features corresponding to the highest level are used as the multi-level airspace event features; The trajectory prediction model is trained through the following steps: Obtain a flight training sample set, which includes labeled flight route parameters; The trajectory prediction model is used to process the sample dynamic optimization prompt text and output the predicted flight path parameters. The parameters of the trajectory prediction model are adjusted based on the difference between the predicted flight path parameters and the labeled flight path parameters; Before processing the sample dynamic optimization prompt text using the trajectory prediction model, the method further includes: Obtain flight pre-training samples without labeled flight status parameters; The flight pre-training samples are subjected to comparative learning feature encoding to generate flight comparative feature vectors; The flight comparison feature vector is processed by the initial trajectory prediction model to output predicted flight feature parameters; The model pre-training loss value is determined based on the predicted flight feature parameters and the flight comparison feature vector; Based on the model pre-training loss value, the parameters of the initial trajectory prediction model are optimized to obtain a pre-trained trajectory prediction model; The step of performing contrastive learning feature encoding on the flight pre-training samples to generate flight contrastive feature vectors includes: Identify airspace topology patterns in the flight pre-training samples; Match the core flight feature encoding corresponding to the airspace topology pattern; Generate a comparative feature code for the core flight feature code; The core flight feature encoding and the contrast feature encoding are coupled to generate the flight contrast feature vector.

2. The flight route dynamic optimization method based on air traffic control data according to claim 1, characterized in that, The step of performing flight parameter feature transformation on the airspace event entity to be processed to generate an airspace event feature vector at the current level includes: Obtain the spatial event description text corresponding to the spatial event entity to be processed; The spatial event description text is processed by a pre-set spatial semantic encoder to output an event semantic feature vector; The event semantic feature vector is transformed by flight parameter dimension to generate the airspace event feature vector of the current level.

3. The flight route dynamic optimization method based on air traffic control data according to claim 1, characterized in that, The step of performing feature fusion processing based on the airspace event entity to be processed at the next level and the airspace event feature vector at the current level to generate target updated airspace event features includes: The mean value of the spatial event feature vectors corresponding to all spatial event entities to be processed in the next level is calculated to generate secondary spatial event features. The secondary airspace event features and the updated airspace event features are weighted and fused to generate the target updated airspace event features.

4. The flight route dynamic optimization method based on air traffic control data according to claim 1, characterized in that, The step of determining a set of similar flight status parameters based on the similarity between the updated flight status parameters and the historical flight status parameters corresponding to other aircraft includes: Calculate the airspace state similarity between the updated flight situation parameters and each historical flight situation parameter; Filter specific historical flight status parameters whose airspace state similarity exceeds a preset similarity threshold; The historical flight path parameters corresponding to the specific historical flight situation parameters are added to the set of similar flight situation parameters.

5. The flight route dynamic optimization method based on air traffic control data according to claim 1, characterized in that, The step of generating dynamic optimization prompt text based on the updated flight situation parameters and the set of similar flight situation parameters includes: The current flight constraints are determined based on the updated flight status parameters. Extract similar flight path trajectory data from the set of similar flight situation parameters; The current flight constraints and the similar flight path trajectory data are combined to synthesize path description text, generating the dynamic optimization prompt text.

6. The flight route dynamic optimization method based on air traffic control data according to claim 1, characterized in that, Before acquiring real-time air traffic control data, the method further includes: Constructing a multi-level airspace topology network includes: Acquire historical airspace operation data, which includes aircraft trajectory instances, airspace control event instances, and flight conflict instances; Determine the airspace topology relationships between the aircraft trajectory instance, the airspace control event instance, and the flight conflict instance; The multi-level spatial topology network is generated based on the spatial topology relationship.

Images