Flight delay prediction method and system based on digital twinning

Abstract

The invention discloses a flight delay prediction method and system based on digital twinning, and relates to the technical field of aviation services, and the method comprises the steps: constructing a multi-source heterogeneous data fused digital twinning body comprising an airport operation state, an aircraft health state and airspace dynamic traffic; accessing data of multiple dimensions in real time through the digital twinborn body; wherein the plurality of dimensions comprise weather, airspace and guarantee resources; the method comprises the following steps: carrying out deep learning training on historical delay data by utilizing a long short-term memory network, extracting time sequence characteristics of flight delay, and capturing a delay propagation rule and multi-node interaction influence in an aviation network in combination with multi-dimensional data accessed in real time so as to predict flight delay probability and influence range in a future time period.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of aviation services, in particular to a flight delay prediction method and system based on digital twinning. BACKGROUND

[0002] In the field of air transportation, flight delays are a long-standing and difficult-to-avoid problem. Traditional flight delay prediction methods, such as statistical analysis based on historical data, while providing prediction results to some extent, are often limited in accuracy and timeliness when faced with complex and variable flight operation environments.

[0003] With the rapid development of the aviation transportation industry, flight delay problems have become a core challenge affecting the efficiency of civil aviation operations and the experience of passengers. It not only affects the operational efficiency of airlines, but also reduces the travel experience of passengers. According to statistics, the flight delay rate caused by factors such as adverse weather, mechanical failure, air space resource conflict and multi-airport ripple effect is as high as 20%-30% globally, resulting in huge economic losses and even causing a chain of aviation network paralysis.

[0004] Traditional delay prediction methods are mostly based on statistical models, such as regression analysis models, to predict flight delays. However, most statistical models rely on static historical data for training and cannot access real-time information such as aircraft sensor data and air traffic control dynamic instructions in real time, resulting in prediction lag; at the same time, the model's generalization ability significantly decreases when faced with extreme weather, sudden failures and other non-steady-state scenarios.

[0005] In view of the above problems, no effective solution has been proposed so far. SUMMARY

[0006] The embodiments of the present application provide a flight delay prediction method and system based on digital twinning to solve the above technical problems.

[0007] The application provides a flight delay prediction method based on digital twinning, comprising: constructing a digital twin containing multi-source heterogeneous data fusion of airport operation state, aircraft health state and air space dynamic flow; Real-time access to multiple dimensions of data through the digital twin; wherein the multiple dimensions include meteorology, air space and support resources; Using a long short-term memory network to perform deep learning training on historical delay data to extract time series features of flight delays, combining real-time access to multiple dimensions of data, capturing delay propagation rules and multi-node interaction effects in the aviation network to predict flight delay probability and impact range in the future period.

[0008] Furthermore, the construction of a digital twin that integrates multi-source heterogeneous data, including airport operational status, aircraft health status, and dynamic airspace traffic, includes: Based on the cross-domain entity relationship modeling strategy, entity alignment and relationship extraction are performed on the facility load rate in airport operation status, sensor time-series signals in aircraft health status, and waypoint congestion index in airspace dynamic traffic to construct a cross-domain spatiotemporal correlation matrix. Based on the cross-domain spatiotemporal correlation matrix, a data-driven simulation scenario for the digital twin is dynamically generated.

[0009] Furthermore, the method of capturing the propagation patterns of delays and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in future periods includes: A high-order graph attention network is constructed. Through the multi-dimensional data accessed in real time by the digital twin, the airport is abstracted as a network node, and an aviation high-order hypergraph is constructed with flight take-off and landing relationships and shared aircraft scheduling links as hyperedges. The multi-airport collaborative delay coefficient is calculated by using a hyperedge convolution module, and the cascading diffusion path of delay in the hyperedge topology is dynamically inferred by combining the time series features output by the long short-term memory network. Based on the cascading diffusion path, the probability of flight delays and the scope of impact in the future are predicted.

[0010] Furthermore, in the construction of the aeronautical high-order hypergraph, a dynamic pruning strategy is adopted. If the flight scheduling frequency of a certain hyperedge is lower than the critical value within N consecutive time windows, the hyperedge is marked as a low-activity state and its weight is reduced in the delay prediction. At the same time, newly emerging temporary diversion flight links are monitored in real time and included as temporary hyperedges in the topology update. Here, N is a positive integer.

[0011] Furthermore, the step of calculating the multi-airport collaborative delay coefficient through the hyperedge convolution module and dynamically inferring the cascading diffusion path of delays in the hyperedge topology by combining the time series features output by the long short-term memory network includes: Calculate the multi-airport collaborative delay coefficient ; Based on the delay diffusion kinetics equation, the cascade diffusion path of delay in the hyperedge topology is dynamically predicted; the delay diffusion kinetics equation is as follows.

[0012] Furthermore, the prediction of the probability and scope of flight delays in the future based on the cascading diffusion path includes: The hyperedge dynamic sensing algorithm is executed, and based on the cascading diffusion path and the multi-airport collaborative delay coefficient, the following operations are performed: the ground staff saturation and parking space occupancy rate of the airports associated with the hyperedge are scanned. If the resource shortage rate is greater than the preset shortage rate for three consecutive monitoring periods, the hyperedge is marked as a vulnerable link. An exponential decay is applied to the delay propagation weight of the vulnerable link, and the decay rate is positively correlated with the rate of weather deterioration. The bidirectional propagation verification algorithm is executed. Along the hyperedge topology direction, the calculated minimum energy path is used as the baseline propagation trajectory, and the real-time flight scheduling density is superimposed to accumulate the delay probability. The delay probability is accumulated by tracing back from the downstream airport to the source of the delay, identifying the cross-airport chain reaction caused by shared aircraft scheduling, and correcting the overfitting error in the forward inference. By combining the outputs of the hyperedge dynamic perception algorithm and the bidirectional propagation verification algorithm, predictions are generated for the probability and impact range of flight delays in future time periods.

[0013] Furthermore, the data-driven simulation scenario for dynamically generating the digital twin includes: Based on the federated learning framework, edge computing nodes are deployed on local servers in airlines, air traffic control departments, and airports to transmit entity relationship subgraph features through homomorphic encryption. After aggregating encrypted features on the central server, Gaussian noise is added using differential privacy technology to generate a global simulation scenario and feed it back to all participants, thus achieving collaborative modeling under data privacy protection.

[0014] Furthermore, the cross-domain entity relationship modeling strategy includes: A unique entity identifier is constructed based on airport facility code, aircraft tail number and waypoint ID, and entity attributes from different data sources are matched using a spatiotemporal alignment algorithm; By using the TransR relational embedding model, the physical and logical relationships between entities are mapped to a low-dimensional vector space, generating a dynamically updatable entity relationship graph.

[0015] Furthermore, after capturing the propagation patterns of delays and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in future periods, the method further includes: The predicted probability of flight delays in the future is mapped onto a 3D geographic information twin platform, and the predicted impact range is displayed in the form of a heat map; When the delay probabilities of multiple related airports form a closed propagation loop, an early warning signal is triggered, and a decision tree simulation report containing rerouting suggestions and ground resource allocation plans is generated.

[0016] This application provides a flight delay prediction system based on digital twins, including: a digital twin construction module for constructing a digital twin that incorporates multi-source heterogeneous data fusion, including airport operation status, aircraft health status, and airspace dynamic traffic; A multi-dimensional data access module is used to access data from multiple dimensions in real time through the digital twin; these multiple dimensions include meteorology, airspace, and support resources. The delay prediction module uses a long short-term memory network to train deep learning on historical delay data, extracts the time series features of flight delays, and combines real-time access to data from multiple dimensions to capture the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network, in order to predict the probability and scope of flight delays in the future.

[0017] Based on the embodiments provided in this application, a digital twin is constructed by fusing multi-source heterogeneous data, including airport operational status, aircraft health status, and dynamic airspace traffic. This data is then integrated in real-time with data from multiple dimensions, including meteorology, airspace, and support resources. This breaks the limitations of traditional prediction methods that rely solely on static historical data, enabling a more comprehensive and real-time capture of various factors influencing flight delays and providing a richer and more accurate data foundation for prediction. Furthermore, deep learning training of historical delay data using a Long Short-Term Memory (LSTM) network deeply extracts the time-series features of flight delays, which, compared to traditional statistical models, better captures long-term dependencies and complex patterns in the data.

[0018] Simultaneously, by combining real-time access to multi-dimensional data, the system accurately captures the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network. This allows for more accurate predictions of the probability and scope of flight delays in future periods, effectively addressing the poor generalization ability of traditional methods in non-steady-state scenarios. Especially under complex conditions such as extreme weather and sudden malfunctions, it maintains high prediction accuracy. Through real-time data access via a digital twin, it can promptly acquire real-time information such as aircraft sensor data and dynamic air traffic control instructions, overcoming the prediction lag problem of traditional methods. This enables real-time, dynamic prediction of flight delays, providing airlines, air traffic control departments, and passengers with more timely decision-making support. It helps in taking proactive measures to address flight delays, improving aviation operational efficiency and passenger travel experience.

[0019] This application applies to all aspects of the air transport sector and is of great significance for airlines to optimize flight schedules, rationally allocate capacity, and improve operational efficiency. Air traffic control departments can adjust airspace resource allocation in advance based on forecast results to alleviate airspace congestion and reduce the chain reaction caused by flight delays. Passengers can also receive flight delay information earlier and make reasonable travel arrangements, thereby comprehensively improving the overall efficiency and service quality of the air transport system. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of an optional digital twin-based flight delay prediction method according to an embodiment of this application; Figure 2 A flowchart illustrating another optional digital twin-based flight delay prediction method according to an embodiment of this application; Figure 3 This is a structural diagram of an optional digital twin-based flight delay prediction system according to an embodiment of this application.

[0021] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Digital twin technology, as an emerging digital solution, has demonstrated strong application potential across multiple industries. By establishing a mapping relationship between physical entities and virtual models, this technology enables real-time perception, dynamic simulation, and intelligent prediction of the physical world. Applying this technology to flight delay prediction is expected to achieve comprehensive and in-depth analysis of the flight operation environment, thereby improving the accuracy and timeliness of predictions.

[0024] By integrating multi-dimensional data related to flight operations, such as meteorological data, airport facility status, and air traffic control information, digital twin technology can construct a virtual flight operation environment model. This model can reflect the dynamic changes in the flight operation environment in real time, providing a richer information foundation for flight delay prediction. Furthermore, based on this model, intelligent early warning and decision support can be implemented, providing relevant departments and passengers with more timely and effective response measures.

[0025] Optionally, such as Figure 1 As shown, this application provides a flight delay prediction method based on digital twins, including: S101, construct a digital twin that integrates multi-source heterogeneous data including airport operation status, aircraft health status, and dynamic airspace traffic; S102 accesses data from multiple dimensions in real time through a digital twin; these dimensions include meteorology, airspace, and support resources. S103 utilizes a long short-term memory network to train deep learning on historical delay data, extracts the time-series features of flight delays, and combines real-time access to data from multiple dimensions to capture the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network, in order to predict the probability and scope of flight delays in the future.

[0026] Based on the embodiments provided in this application, a digital twin is constructed by fusing multi-source heterogeneous data, including airport operational status, aircraft health status, and dynamic airspace traffic. This data is then integrated in real-time with data from multiple dimensions, including meteorology, airspace, and support resources. This breaks the limitations of traditional prediction methods that rely solely on static historical data, enabling a more comprehensive and real-time capture of various factors influencing flight delays and providing a richer and more accurate data foundation for prediction. Furthermore, deep learning training of historical delay data using a Long Short-Term Memory (LSTM) network deeply extracts the time-series features of flight delays, which, compared to traditional statistical models, better captures long-term dependencies and complex patterns in the data.

[0027] Simultaneously, by combining real-time access to multi-dimensional data, the system accurately captures the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network. This allows for more accurate predictions of the probability and scope of flight delays in future periods, effectively addressing the poor generalization ability of traditional methods in non-steady-state scenarios. Especially under complex conditions such as extreme weather and sudden malfunctions, it maintains high prediction accuracy. Through real-time data access via a digital twin, it can promptly acquire real-time information such as aircraft sensor data and dynamic air traffic control instructions, overcoming the prediction lag problem of traditional methods. This enables real-time, dynamic prediction of flight delays, providing airlines, air traffic control departments, and passengers with more timely decision-making support. It helps in taking proactive measures to address flight delays, improving aviation operational efficiency and passenger travel experience.

[0028] This application applies to all aspects of the air transport sector and is of great significance for airlines to optimize flight schedules, rationally allocate capacity, and improve operational efficiency. Air traffic control departments can adjust airspace resource allocation in advance based on forecast results to alleviate airspace congestion and reduce the chain reaction caused by flight delays. Passengers can also receive flight delay information earlier and make reasonable travel arrangements, thereby comprehensively improving the overall efficiency and service quality of the air transport system.

[0029] Furthermore, a digital twin is constructed that integrates multi-source heterogeneous data, including airport operational status, aircraft health status, and dynamic airspace traffic, comprising: Based on the cross-domain entity relationship modeling strategy, entity alignment and relationship extraction are performed on the facility load rate in airport operation status, sensor time-series signals in aircraft health status, and waypoint congestion index in airspace dynamic traffic to construct a cross-domain spatiotemporal correlation matrix. Among them, the cross-domain entity relationship modeling strategy refers to a set of technologies that map entities and their interaction relationships from heterogeneous data sources such as airport facilities, aircraft, and airspace units to a unified representation space through methods such as encoding conversion, spatiotemporal alignment, and relationship embedding. Entity alignment includes mapping airport facility codes, aircraft tail numbers, and waypoint IDs to a unified spatiotemporal coordinate system. The cross-domain spatiotemporal association matrix is ​​used to solve the problems of data modality differences and timestamp asynchrony. A data-driven simulation scenario that dynamically generates digital twins based on a cross-domain spatiotemporal correlation matrix.

[0030] Furthermore, such as Figure 2 As shown, this study captures the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in future time periods, including: S201 constructs a high-order graph attention network, which abstracts the airport into network nodes through real-time access to multi-dimensional data from digital twins, and constructs an aviation high-order hypergraph with flight take-off and landing relationships and shared aircraft scheduling links as hyperedges; S202 calculates the multi-airport collaborative delay coefficient through the hyperedge convolution module and, combined with the time series features output by the long short-term memory network, dynamically infers the cascading diffusion path of delays in the hyperedge topology. S203, based on the cascading diffusion path, predicts the probability and scope of flight delays in the future.

[0031] In the embodiments of this application, aerospace high-order hypergraph ; Among them, node set Represents a set of airports, each node The embedding dimension is [longitude, latitude, instantaneous capacity]; hyperedge set This indicates the shared machine model scheduling link, and each superedge It contains at least 3 nodes, whose weights are initialized by the historical delay propagation probability; Furthermore, in the construction of the aviation high-order hypergraph, a dynamic pruning strategy is adopted. If the flight scheduling frequency of a certain hyperedge is lower than the critical value within N consecutive time windows, the hyperedge is marked as a low-activity state and its weight is reduced in the delay prediction. At the same time, newly emerging temporary diversion flight links are monitored in real time and included as temporary hyperedges in the topology update. Here, N is a positive integer.

[0032] Furthermore, the multi-airport collaborative delay coefficient is calculated using a hyperedge convolution module, and the time-series features output by the long short-term memory network are combined to dynamically infer the cascading diffusion path of delays in the hyperedge topology, including: The multi-airport collaborative delay coefficient is calculated based on the following formula. ;

[0033] in, All are trainable parameter matrices, with dimensions... Attention weights between nodes are learned through backpropagation; Indicates the superedge The topological neighbor node refers to the node that is adjacent to the hyperedge. Related airports that have flight operations with any airport in China; For super-edge The member nodes in the list represent the specific airports participating in the shared aircraft type scheduling link; for example, if the super-edge This indicates the "scheduling link of B111-800 aircraft in Pudong (ZSPD) - Capital (ZBAA) - Baiyun (ZGGG)". These can be three airport nodes: ZSPD, ZBAA, and ZGGG. For super-edge Topological neighboring nodes, ; The target node time series feature vector extracted for the Long Short-Term Memory Network includes the following features: historical delay trend: exponentially weighted moving average of delay rate (5-minute granularity) over the past 4 hours; real-time weather change: predicted rate of change of visibility, wind speed, and precipitation over the next 30 minutes; capacity fluctuation: 15-minute variance of gate / runway load rate. Represents the contextual temporal feature vector of neighboring nodes; This represents the feature dimension, which is fixed at 128, corresponding to the number of hidden layer neurons in the Long Short-Term Memory network. This is an activation function used to filter negatively propagating signals; Based on the delay diffusion kinetics equation, the cascade diffusion path of delay in the hyperedge topology is dynamically predicted; the delay diffusion kinetics equation is:

[0034] in, Represents the delay intensity matrix. This indicates the delay impact of airport i on airport j; This represents a time variable used to describe how the intensity of the delay changes over time. This represents the hypergraph gradient operator, used to calculate the delay propagation potential on the topological structure; This represents the multi-airport collaborative delay coefficient; This represents a real-time resource allocation matrix, which is dynamically updated from resource availability data. , All are adaptive balance factors; ; .

[0035] Based on the embodiments provided in this application, by combining the time-series features output by the hyperedge convolution module with the long short-term memory network, the collaborative delay coefficient of multiple airports can be accurately captured. Then, based on the delay diffusion dynamics equation, the cascading diffusion path of delays in the hyperedge topology can be dynamically inferred. This fully considers the complex relationships between airports and the impact of real-time resource allocation on delay diffusion. Compared to traditional prediction methods based solely on time-series analysis or fixed topology structures, this method can more accurately predict the probability and scope of flight delays in future periods, effectively improving the accuracy and reliability of flight delay prediction. This provides airlines and airport operations control departments with more precise decision-making support, helping to formulate more reasonable flight scheduling and resource allocation strategies in advance, reducing the negative impact of flight delays on the overall operational efficiency of air transport, and improving the quality of air transport services and passenger satisfaction.

[0036] Furthermore, based on the cascading diffusion path, the probability and scope of flight delays in the future are predicted, including: The hyperedge dynamic sensing algorithm is executed, and based on the cascading diffusion path and the multi-airport collaborative delay coefficient, the following operations are performed: the ground staff saturation and parking space occupancy rate of the airports associated with the hyperedge are scanned. If the resource shortage rate is greater than the preset shortage rate for three consecutive monitoring periods, the hyperedge is marked as a vulnerable link. An exponential decay is applied to the delay propagation weight of the vulnerable link, and the decay rate is positively correlated with the rate of weather deterioration. The bidirectional propagation verification algorithm is executed. Along the hyperedge topology direction, the calculated minimum energy path is used as the baseline propagation trajectory, and the real-time flight scheduling density is superimposed to accumulate the delay probability. The delay probability is accumulated by tracing back from the downstream airport to the source of the delay, identifying the cross-airport chain reaction caused by shared aircraft scheduling, and correcting the overfitting error in the forward inference. By combining the outputs of the hyperedge dynamic sensing algorithm and the bidirectional propagation verification algorithm, a prediction of the probability and impact range of flight delays in future time periods is generated. The prediction results are then injected into the 3D visualization engine of the digital twin. When more than two vulnerable links are detected to form a closed loop, an optimized runway allocation plan is automatically pushed to the air traffic control terminal.

[0037] Delay probability determination: When the probability deviation of an airport in two-way propagation is <15%, the average value is taken as the final value; if the deviation is ≥15%, a manual review mechanism is triggered. Impact area delineation: Centered on the target airport, the area is expanded in a fan shape based on the airway structure, prioritizing adjacent airports with alternate airport capacity of <50%; The hyperedge dynamic perception algorithm provided in this application can monitor the resource status of hyperedge-related airports in real time, such as ground staff saturation and parking space occupancy, promptly mark vulnerable links, and reasonably attenuate the delay propagation weight, effectively considering the aggravating effect of airport resource constraints on delay propagation. The bidirectional propagation verification algorithm verifies and corrects the delay propagation path from both forward and backward propagation directions, avoiding the error accumulation and overfitting problems that may occur with single-direction prediction. Compared with conventional unidirectional prediction methods, this method can generate more comprehensive and accurate predictions of the probability and impact range of future flight delays when facing flight traffic fluctuations caused by seasonal changes or temporary flight increases / decreases. This significantly improves the robustness and adaptability of flight delay prediction, enabling relevant departments to more efficiently coordinate responses to flight delay issues, adjust flight plans in a timely manner, rationally allocate support resources, effectively reduce the chain reaction of flight delays, improve the overall operational support capacity and emergency response level of airports, and reduce economic losses and social impacts caused by flight delays.

[0038] Furthermore, data-driven simulation scenarios that dynamically generate digital twins include: Based on the federated learning framework, edge computing nodes are deployed on local servers in airlines, air traffic control departments, and airports to transmit entity relationship subgraph features through homomorphic encryption. After aggregating encrypted features on the central server, Gaussian noise is added using differential privacy technology to generate a global simulation scenario and feed it back to all participants, thus achieving collaborative modeling under data privacy protection.

[0039] Furthermore, cross-domain entity relationship modeling strategies include: A unique entity identifier is constructed based on airport facility code, aircraft tail number and waypoint ID, and entity attributes from different data sources are matched using a spatiotemporal alignment algorithm; By using the TransR relational embedding model, the physical and logical relationships between entities are mapped to a low-dimensional vector space, generating a dynamically updatable entity relationship graph.

[0040] In the training process of the TransR relation embedding model, the entity space dimension is set to 128 and the relation space dimension to 64, and the flight punctuality rate is used as the initialization parameter for relation weights. The physical relationships between entities are represented, such as aircraft-airline; the logical relationships are represented, such as flight-affected subsequent flights. Furthermore, after capturing the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in future periods, the method also includes: The predicted probability of flight delays in the future is mapped onto a 3D geographic information twin platform, and the predicted impact range is displayed in the form of a heat map; When the delay probabilities of multiple related airports form a closed propagation loop, an early warning signal is triggered, and a decision tree simulation report containing rerouting suggestions and ground resource allocation plans is generated.

[0041] like Figure 3 As shown, this application provides a flight delay prediction system based on digital twins, including: Digital twin construction module 301 is used to construct a digital twin that incorporates multi-source heterogeneous data fusion, including airport operation status, aircraft health status, and dynamic airspace traffic. The multi-dimensional data access module 302 is used to access data from multiple dimensions in real time through a digital twin; these multiple dimensions include meteorology, airspace, and support resources. The delay prediction module 303 is used to train historical delay data using a long short-term memory network to extract the time series features of flight delays. Combined with real-time access to data from multiple dimensions, it captures the delay propagation patterns and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in the future.

[0042] It should be noted that the embodiments implemented on the flight delay prediction system side based on digital twins in this application can be referenced with the embodiments implemented on the flight delay prediction method side based on digital twins, and will not be described in detail here.

[0043] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting flight delays based on digital twins, characterized in that, include: Construct a digital twin that integrates multi-source heterogeneous data, including airport operational status, aircraft health status, and dynamic airspace traffic. The digital twin provides real-time access to data from multiple dimensions, including meteorology, airspace, and support resources. By using a long short-term memory network to train deep learning on historical delay data, the time series features of flight delays are extracted. Combined with real-time access to multi-dimensional data, the propagation patterns of delays and the interactive effects of multiple nodes in the aviation network are captured to predict the probability and scope of flight delays in the future.

2. The flight delay prediction method based on digital twins according to claim 1, characterized in that, The construction of a digital twin, which integrates multi-source heterogeneous data including airport operational status, aircraft health status, and dynamic airspace traffic, includes: Based on the cross-domain entity relationship modeling strategy, entity alignment and relationship extraction are performed on the facility load rate in airport operation status, sensor time-series signals in aircraft health status, and waypoint congestion index in airspace dynamic traffic to construct a cross-domain spatiotemporal correlation matrix. Based on the cross-domain spatiotemporal correlation matrix, a data-driven simulation scenario for the digital twin is dynamically generated.

3. The flight delay prediction method based on digital twins according to claim 1, characterized in that, The method of capturing delay propagation patterns and multi-node interactions in the aviation network to predict the probability and scope of flight delays in future periods includes: A high-order graph attention network is constructed. By using multi-dimensional data accessed in real time by the digital twin, the airport is abstracted as a network node, and a high-order hypergraph of aviation is constructed with flight take-off and landing relationships and shared aircraft scheduling links as hyperedges. ; Among them, node set Represents a set of airports, each node The embedding dimension is [longitude, latitude, instantaneous capacity]; hyperedge set This indicates the shared machine model scheduling link, with each superedge... It contains at least 3 nodes, whose weights are initialized by the historical delay propagation probability; The multi-airport collaborative delay coefficient is calculated by using a hyperedge convolution module, and the cascading diffusion path of delay in the hyperedge topology is dynamically inferred by combining the time series features output by the long short-term memory network. Based on the cascading diffusion path, the probability of flight delays and the scope of impact in the future are predicted.

4. The flight delay prediction method based on digital twins according to claim 3, characterized in that, In the construction of the aeronautical high-order hypergraph, a dynamic pruning strategy is adopted. If the flight scheduling frequency of a certain hyperedge is lower than the critical value within N consecutive time windows, the hyperedge is marked as a low-activity state and its weight is reduced in the delay prediction. At the same time, newly emerging temporary diversion flight links are monitored in real time and included as temporary hyperedges in the topology update. Here, N is a positive integer.

5. The flight delay prediction method based on digital twins according to claim 3, characterized in that, The step of calculating the multi-airport collaborative delay coefficient using a hyperedge convolution module and dynamically inferring the cascading diffusion path of delays in the hyperedge topology by combining the time-series features output by the long short-term memory network includes: Calculate the multi-airport collaborative delay coefficient ; Based on the delay diffusion dynamics equation, the cascade diffusion path of delay in the hyperedge topology is dynamically predicted.

6. The flight delay prediction method based on digital twins according to claim 5, characterized in that, The prediction of the probability and scope of flight delays in the future based on the cascading diffusion path includes: The hyperedge dynamic sensing algorithm is executed, and based on the cascading diffusion path and the multi-airport collaborative delay coefficient, the following operations are performed: the ground staff saturation and parking space occupancy rate of the airports associated with the hyperedge are scanned. If the resource shortage rate is greater than the preset shortage rate for three consecutive monitoring periods, the hyperedge is marked as a vulnerable link. An exponential decay is applied to the delay propagation weight of the vulnerable link, and the decay rate is positively correlated with the rate of weather deterioration. The bidirectional propagation verification algorithm is executed. Along the hyperedge topology direction, the calculated minimum energy path is used as the baseline propagation trajectory, and the real-time flight scheduling density is superimposed to accumulate the delay probability. The delay probability is accumulated by tracing back from the downstream airport to the source of the delay, identifying the cross-airport chain reaction caused by shared aircraft scheduling, and correcting the overfitting error in the forward inference. By combining the outputs of the hyperedge dynamic perception algorithm and the bidirectional propagation verification algorithm, predictions are generated for the probability and impact range of flight delays in future time periods.

7. The flight delay prediction method based on digital twins according to claim 2, characterized in that, The data-driven simulation scenario for dynamically generating the digital twin includes: Based on the federated learning framework, edge computing nodes are deployed on local servers in airlines, air traffic control departments, and airports to transmit entity relationship subgraph features through homomorphic encryption. After aggregating encrypted features on the central server, Gaussian noise is added using differential privacy technology to generate a global simulation scenario and feed it back to all participants, thus achieving collaborative modeling under data privacy protection.

8. The flight delay prediction method based on digital twins according to claim 2 or 7, characterized in that, The cross-domain entity relationship modeling strategy includes: A unique entity identifier is constructed based on airport facility code, aircraft tail number and waypoint ID, and entity attributes from different data sources are matched using a spatiotemporal alignment algorithm; By using the TransR relational embedding model, the physical and logical relationships between entities are mapped to a low-dimensional vector space, generating a dynamically updatable entity relationship graph.

9. The flight delay prediction method based on digital twins according to claim 1, characterized in that, After capturing the propagation patterns of delays and the interactive effects of multiple nodes in the aviation network to predict the probability and scope of flight delays in future periods, the method further includes: The predicted probability of flight delays in the future is mapped onto a 3D geographic information twin platform, and the predicted impact range is displayed in the form of a heat map; When the delay probabilities of multiple related airports form a closed propagation loop, an early warning signal is triggered, and a decision tree simulation report containing rerouting suggestions and ground resource allocation plans is generated.

10. A flight delay prediction system based on digital twins, the system implementing the method as described in claim 1, characterized in that, include: The digital twin construction module is used to build a digital twin that integrates multi-source heterogeneous data, including airport operation status, aircraft health status, and dynamic airspace traffic. A multi-dimensional data access module is used to access data from multiple dimensions in real time through the digital twin; these multiple dimensions include meteorology, airspace, and support resources. The delay prediction module uses a long short-term memory network to train deep learning on historical delay data, extracts the time series features of flight delays, and combines real-time access to data from multiple dimensions to capture the patterns of delay propagation and the interactive effects of multiple nodes in the aviation network, in order to predict the probability and scope of flight delays in the future.

Images