An airport cluster runway operation mode prediction method based on dynamic graph

Abstract

The application relates to an airport cluster runway operation mode prediction method based on a dynamic graph, and belongs to the technical field of air traffic control operation management. The application constructs an airport cluster runway operation mode prediction dynamic graph model, obtains the prediction of a multi-runway operation mode based on the runway operation mode prediction dynamic graph model, is suitable for an airport cluster operation scene, continuously predicts the multi-runway operation mode of all operation time periods, not only supports accurate runway scheduling of air traffic controllers, but also can provide strong data support for the overall strategy of the airport cluster.

Description

Technical Field

[0001] This invention relates to the field of air traffic control operation management technology, specifically to a method for predicting the runway operation mode of airport groups based on dynamic graphs. Background Technology

[0002] With the rapid growth of air traffic, predicting runway operation patterns for airport clusters has become a crucial element in ensuring the safe and efficient operation of airports. In the context of globalization, airports are no longer nearly isolated entities, but rather collectively form a complex airport cluster system. This shift means that the operation pattern of a single airport will directly or indirectly affect the operations of other airports within the cluster. For example, delays at one airport can trigger a chain reaction in aircraft scheduling across the entire cluster.

[0003] Runway operation patterns refer to the regulations and arrangements for runways based on aircraft takeoffs, landings, and other operations within a given time period. Reasonable forecasting of runway operation patterns within an airport cluster ensures that aircraft have appropriate time intervals and optimal conditions for takeoff and landing on specific airport runways (runways selected for specific circumstances at different airports). Factors influencing runway operation pattern forecasting within an airport cluster include the airport's geographical location and layout, current weather conditions, air traffic flow, runway physical condition, flight type and destination, and safety requirements.

[0004] Meanwhile, to meet the ever-increasing demand for air travel, airports must operate efficiently. Optimizing runway utilization by adjusting and adopting different runway operation modes allows airports to handle more aircraft, thereby increasing their throughput. Furthermore, proper runway management can reduce flight delays caused by runway congestion or other reasons, reduce noise and other environmental impacts on the surrounding area, ensure maximum resource utilization, and better coordinate aircraft traffic and respond to various emergencies. Therefore, runway operation modes are crucial for the safe and efficient operation of airports. Different airports may choose runway operation modes based on their specific needs and constraints, and the selection and use of runways are also influenced by air traffic control, weather conditions, equipment availability, and other factors.

[0005] In most cases, runway operation patterns are not predicted at the technical level, but are mainly changed based on fixed rules and control experience. However, this can easily lead to airport congestion or waste of runway resources. To improve airport operational efficiency, runway operation pattern prediction technologies have emerged. However, existing technologies only consider predictions at a single moment and use relatively simple data, ignoring complex interactions, resulting in inaccurate predictions. More importantly, these methods usually only consider the situation of a single airport, ignoring the coupling relationships between airports in an airport cluster, which often leads to inefficient regional operations. Summary of the Invention

[0006] In view of the above problems, the present invention provides a method for predicting runway operation modes of airport clusters based on dynamic graphs. A dynamic graph model of multi-runway operation modes of airport clusters is constructed, and predictions of multi-runway operation modes are obtained based on the dynamic graph model. This method is suitable for airport cluster operation scenarios and continuously predicts multi-runway operation modes for all operation periods. It not only supports air traffic controllers to perform precise runway scheduling, but also provides strong data support for the overall strategy of airport clusters.

[0007] This invention provides a method for predicting airport group runway operation modes based on dynamic graphs, comprising:

[0008] Step 1: Based on historical data of the airport cluster, obtain the operation modes of multiple runways in the airport cluster and their corresponding influencing factors, and construct the node feature matrix of multiple runway operation modes;

[0009] Preferably, the specific steps for constructing the node feature matrix in step 1 include:

[0010] Using the runway operation mode i as a node, obtain the influencing factors corresponding to the runway operation mode, and establish node features v based on the influencing factors. i ;

[0011] υ i =[υ i1 ,υ i2 ,...,υ iNUM ]

[0012] Among them, υ inum Let num be the influence value of the node corresponding to runway operation mode i and the num-th influencing factor, where num = 1, 2, 3…NUM, NUM is the total number of influencing factors, and i = 1, 2, 3…N, where N represents the total number of runway operation modes.

[0013] Construct a node feature matrix for multiple runway operation modes.

[0014] Furthermore, the influencing factors corresponding to the runway operation mode include: runway operation mode usage frequency, average duration of runway operation mode, correlation between runway operation modes, weather sensitivity, flight traffic sensitivity, airport cluster effect and / or sensitivity to triggering airport rule events.

[0015] The expression for the correlation between the runway operation modes is:

[0016]

[0017] Among them, R ij u represents the correlation between runway operation mode i and runway operation mode j. ijThe number of times runway operation mode i is transformed into runway operation mode j; i, j∈N, i≠j, N represents the total number of runway operation modes; T is the total time period for runway operation mode prediction.

[0018] The runway operation mode weather sensitivity measure is the strength or preference of the association between a specific runway operation mode and a specific weather condition.

[0019] The flight traffic sensitivity refers to the frequency of use of runway operation mode i under flight traffic conditions; for example, the frequency of use of runway operation mode i when flight traffic is high, and the frequency of use of runway operation mode i when flight traffic is low.

[0020] The airport cluster effect refers to the status and influence of an airport within an airport cluster. For a given airport, its effect within the airport cluster is determined by its traffic correlation and distance from other airports.

[0021] It is understood that the triggering airport rule event sensitivity refers to the frequency of use of runway operation mode i in the event of an unsafe event in the airport group; the unsafe event includes loss of control in the air, runway safety, and controlled flight into the ground.

[0022] Step 2: Treat the mode transition between the two runway operation modes as an edge, and obtain the influencing factors corresponding to the mode transition; obtain the edge features based on the influencing factors corresponding to the mode transition;

[0023] Construct a set of edge features corresponding to multiple runway operation modes;

[0024] It is understood that the mode switching between the two runway operation modes includes: based on specific conditions (e.g., weather conditions, flight traffic volume, flight safety requirements, or specific time periods), in order to ensure the efficient operation of aircraft in the airport group, the runway operation mode i is switched to the adjacent runway operation mode j.

[0025] The influencing factors corresponding to the mode transition include: the frequency of the transition between the two runway operation modes, the average transition delay, the correlation between the frequency of the transition between the two runway operation modes, the correlation between the transition and the weather, the flight traffic during the transition, the airport cluster effect of the transition and / or the airport rule events triggered by the transition.

[0026] The correlation of the switching frequency of the two runway operation modes is: the probability of switching from one runway operation mode to another and then to a third runway operation mode. Assuming an airport has three runway operation modes: A, B and C, if we observe that within a specific time period, the airport switches from runway operation mode A to runway operation mode B, and then frequently switches to runway operation mode C, then the correlation of this switching sequence A→B→C will be relatively high.

[0027] The weather sensitivity of the two runway operation modes refers to the impact of specific weather conditions on an airport's switch from one runway operation mode to another; under certain weather conditions, such as rain or fog, an airport may be more inclined to switch from the current mode to a specific, safer mode.

[0028] It is understandable that the airport cluster effect of the conversion of the various runway operation modes is based on the distance and flight traffic between airports in airport cluster g, where runway operation mode i is converted from airport a to airport b, a, b∈g, and g represents an airport cluster consisting of multiple airports.

[0029] Step 3: Obtain the feature set based on the node feature matrix and the edge feature set;

[0030] A dynamic graph model is constructed, and the feature set and the event-driven time model are embedded into the dynamic graph model to obtain a time-series dynamic graph model;

[0031] Use a time-series dynamic graph model to obtain the time-series message memory of multiple runway operation modes at the current moment;

[0032] The time-series message memory of multiple runway operation modes at the current moment is used to build a dynamic graph model for predicting the runway operation mode;

[0033] Preferably, acquiring the time-series message memory of multiple runway operation modes at the current moment specifically includes:

[0034] A dynamic graph model is constructed, and the feature set and the event-driven time model are embedded into the dynamic graph model to obtain a time-series dynamic graph model;

[0035] Determine whether node i in runway operation mode at the current time t is an independent node;

[0036] It is understandable that at the current time t, runway operation mode i is an independent node representing whether the next runway operation mode predicted by runway operation mode i is still runway operation mode i.

[0037] Runway operation mode i is not an independent node, which means that runway operation mode i has neighboring nodes. Runway operation mode i and its neighboring nodes will switch between runway operation modes.

[0038] If node i in runway operation mode at current time t is an independent node, the feature v of node i in runway operation mode at current time t is obtained based on the node feature matrix. i (t); The current time t runway operation mode i node feature v i (t) Input the time-series dynamic graph model to obtain the runway operation mode node features v before the current time t. iGlobal information s of (t) i (t-), t=1,2,3…T, where T represents the total number of time points;

[0039] It is understandable that the runway operation mode i node feature v before the current time t... i Global information s of (t) i (t-)′ represents the memory corresponding to all events that occur from time 0 to time t;

[0040] Using a graph attention mechanism, the features v of node i representing the runway running mode before the current time t are... i Global information s of (t) i (t-)′ and the current time t, runway operation mode i, node characteristics v i (t) is aggregated to obtain the initial aggregation message m of runway operation mode i at the current time t. i (t)′;

[0041] Obtain multiple events of runway operation mode i before the current time t, and aggregate the message m of runway operation mode i at the current time t based on the multiple events. i (t)′ performs batch aggregation; the multiple events correspond to multiple times, and the times corresponding to the multiple events are in an unordered state, where the time of the latest event is t, and there are a total of B events;

[0042] Obtain the aggregated message of runway operation mode i at the current time t.

[0043] Aggregate the messages of runway operation mode i at the current time t. Features of node i in the runway operation mode before the current time t i Global information s of (t) i (t-)′ performs message propagation and updates to obtain the current time t, runway operation mode i node features v i Global information s of (t) i (t)′;

[0044] The current time t runway operation mode node feature v i Global information s of (t) i (t)′、Runway operation mode i node characteristics v i (t) performs message propagation and generates a time-series message memory Z for the runway operation mode i of the independent node at the current time t. i(t) ′;

[0045] If node i in runway operation mode at current time t is not an independent node, the feature v of node i in runway operation mode at current time t is obtained based on the node feature matrix and the edge feature set. i(t) and its neighboring runway operation mode j node characteristics v j (t), to obtain the edge features e of runway operation mode i and runway operation mode j at the current time t. ij (t); i≠j, i, j∈N, N represents the total number of runway operation modes;

[0046] The runway operation mode i node feature v i (t) and runway operation mode j node characteristics v j (t) Input the time-series dynamic graph model to obtain the runway operation mode i node features v before the current time t. i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) j (t-), Current time t edge feature e ij (t); Using a graph attention mechanism, the features v of node i representing the runway running mode before the current time t are... i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) i (t-), Current time t edge feature e ij (t), current time t, runway operation mode, node characteristics v i (t) is aggregated to obtain the initial aggregation message m of multiple nodes in runway operation mode i at the current time t. i (t); j∈N, N represents the total number of runway operation modes;

[0047] Obtain multiple events of runway operation mode i before time t at the current time t, and based on the multiple events before time t, generate the multi-node initial aggregation message m of runway operation mode i at the current time t. i (t) Perform batch aggregation to obtain the multi-node aggregation message of runway operation mode i at the current time t. It is understandable that the multiple events before the current time t correspond to different times, which is an unordered time state, where the latest event occurs at time t, and there are a total of B events;

[0048] Aggregate the multi-node message of runway operation mode i at the current time t. Features of node i in the runway operation mode before the current time t i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) j(t-) Perform message propagation and update to obtain the current time t, runway operation mode i node features v i Global information s of (t) i (t) and the current time t, runway operation mode, node characteristics v j Global information s j (t);

[0049] The current time t runway operation mode i node feature v i Global information s of (t) i (t) and the current time t, runway operation mode, node characteristics v j Global information s of (t) j (t), current time t edge feature e ij (t), the node characteristics vi(t) of runway operation mode i at current time t, and the node characteristics v of runway operation mode j at current time t. j (t) performs message propagation and generates the time-series message memory Z of the non-independent node runway operation mode i at the current time t. i(t) ;

[0050] By iterating through multiple runway operation modes and repeating the above steps, we can obtain the time sequence message memory of multiple runway operation modes at the current time t.

[0051] Furthermore, the time-series message memory Z of the non-independent node runway operation mode i at the current time t. i(t) The expression is:

[0052]

[0053] Among them, Z i(t) Let `emb(i, t)` be the temporal message memory of runway operation mode i of non-independent node at time `t`; `emb(·)` is the embedding function; `emb(i, t)` is the information collected from the neighbor runway operation modes of runway operation mode i at time `t` through the graph attention mechanism; `Attention(·)` is the graph attention weight, and `v` is the value of `i`. i For the runway operation mode i-node features, v j Let j be the feature of the runway operation mode, i = 1, 2, 3…N. For the process from time 0 to time t, the k-th order neighborhood of runway operation mode i; i, j∈N, N represents the total number of runway operation modes, k=1,2,3…K, K represents the total order of the neighborhood; e ij For edge features; s j (t) represents the node characteristics of runway operation mode j at the current time t. j Global information; s i (t) represents the runway operation mode node v at the current time t. i Global information; vi (t) represents the feature of node i in runway operation mode at the current time t; v j (t) represents the feature of node j in runway operation mode at the current time t; h(·) is the learning function.

[0054] h(s i (t),s j ,(t),e ij ,υ i (t),υ j (t))=α ij ·MLP(s i (t)||s j (t)||e ij ||υ i (t)||υ j (t))

[0055] Where, α ij The attention coefficient is calculated based on the node features of runway operation mode i and runway operation mode j; MLP(·) is a nonlinear function used to learn complex nonlinear representations from the input features.

[0056] The technical solution of this invention obtains the initial aggregated message of node i in the runway operation mode at the current time t based on the graph attention mechanism. This message is used to capture the key features of the node at a specific time and its relationship with other nodes. The initial aggregated messages of the runway operation mode at different time points are aggregated to reflect the dynamic changes of the node over time. While maintaining the time series information of the node, the characteristics of the node at different time points are further integrated. The initial aggregated messages of the runway operation mode at different time points are further aggregated in the time dimension to obtain the aggregated message, which more comprehensively reflects the dynamic characteristics of the node over time. This two-stage aggregation method helps to more accurately capture and understand the behavior of nodes in the time series, thereby providing support for time- and event-based prediction models.

[0057] This invention employs an event-driven mechanism that triggers predictions based on real-time changes in data characteristics. The core of this method lies in monitoring key data indicators and immediately performing predictive analysis when these indicators show significant changes. For example, events that trigger prediction updates include adding a new flight schedule to the system or when sudden weather changes affect runway availability. The dynamic graph model captures these changes in real time and immediately reassesses the operational patterns of the airport complex runways, generating updated predictions. This approach enables the prediction model to respond more sensitively to dynamic changes in actual operations, improving the timeliness and accuracy of predictions, avoiding unnecessary calculations when data is relatively static, thereby optimizing resource utilization and response speed.

[0058] Event-driven forecasting means that model updates are triggered based on the occurrence of events, rather than at fixed time intervals. In this case, events might be: the scheduling or cancellation of new flights; predicted weather events, such as the approach of a storm; unusual flight delays or emergencies; changes in runway approach and exit status. As these events occur, the dynamic graphical model receives new data points and then updates its forecasts based on these data points. This approach is more flexible and adaptable, providing more timely and accurate information for operational decisions.

[0059] Step 4: Based on the runway operation mode prediction dynamic graph model and loss function, predict the runway operation mode at the next time of the current time t runway operation mode i, and obtain the prediction probability of each runway operation mode as the runway operation mode at the next time of the current time t runway operation mode i, thus obtaining multiple prediction probabilities.

[0060] A threshold is set, and the multiple predicted probabilities are compared with the threshold. The runway operation mode that is greater than the threshold is taken as the prediction result of the runway operation mode at the next time step of the current time step t runway operation mode i, so as to obtain multiple prediction results corresponding to the runway operation mode at the next time step of the current time step t runway operation mode i.

[0061] Preferably, when runway operation mode i at the current time t is a non-independent node, the prediction probability expression for the runway operation mode at the next time step of the non-independent node runway operation mode i at the current time t is:

[0062]

[0063] Among them, P iq,t+1 Z represents the predicted probability that runway operation mode i of a non-independent node will be runway operation mode q at the next time t+1; б(·) represents the sigmoid function, i, q∈N, N represents the total number of runway operation modes, and Z q(t) Z is the time-series message memory of the non-independent node runway operation mode q at the current time t; i(t) This is the time-series message memory of the non-independent node runway operation mode i at the current time t, where T is the transpose. Let be the dot product vector of the non-independent node runway operation modes i and q at the current time t.

[0064] The loss function expression for runway operation mode i at the current time t, where it is a non-independent node, is as follows:

[0065]

[0066] Among them, y iq,t+1 Let y represent the link relationship between runway operation mode i and runway operation mode q at time t+1 (non-independent node). iq,t+1This indicates that the link exists, y iq,t+1 =0 indicates that the link does not exist.

[0067] When runway operation mode i is an independent node at the current time t, the prediction probability expression for the runway operation mode at the next time step of the independent node runway operation mode i at the current time t is:

[0068]

[0069] Among them, P′ iq,t+1 Z represents the predicted probability that the runway operation mode i of an independent node at time t+1 is runway operation mode q; б(·) represents the sigmoid function, i, q∈N, N represents the total number of runway operation modes, and Z q(t)′ Z is the time-series message memory of the independent node runway operation mode q at the current time t; i(t)′ This is the time-series message memory of the independent node runway operation mode i at the current time t, where T is the transpose. Let be the dot product vector of the independent node runway operation modes i and q at the current time t.

[0070] The loss function expression for runway operation mode i at the current time t, when i is an independent node, is:

[0071] δ=y′ iq,t+1 log(P′ iq,t+1 )-(1-y′ iq,t+1 log(1-P′) iq,t+1 )

[0072] Among them, y iq,t+1 Let y′ represent the link relationship between independent node runway operation mode i and runway operation mode q at time t+1. iq,t+1 This indicates that the link exists, y′ iq,t+1 =0 indicates that the link does not exist.

[0073] Furthermore, thresholds are set based on historical runway usage data, flight schedule data, environmental and meteorological data, operational constraints and rules, runway condition data, node and link relationship data, and tag data.

[0074] Compared with the prior art, the present invention has at least the following beneficial effects:

[0075] (1) This invention takes into account the complexity and diversity of airport cluster operations. By combining time, network structure within and between airport clusters and airport characteristic information, it can adapt to various airport cluster operation scenarios. In particular, when airport traffic flow or weather conditions change, it can achieve high-precision runway operation mode prediction and provide controllers with accurate runway operation mode suggestions.

[0076] (2) This invention combines the features of nodes and edges to achieve a rapid response to real-time data, helping airport cluster management to formulate more scientific and reasonable operation strategies;

[0077] (3) This invention optimizes runway scheduling, avoids unnecessary delays, and improves the overall operational efficiency and safety of airport clusters.

[0078] (4) The runway operation mode prediction method of the present invention has excellent scalability and can adapt to larger airport clusters or introduce more features and data sources, thereby meeting the ever-evolving operational needs of airport clusters. Attached Figure Description

[0079] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0080] Figure 1 This is a schematic diagram of the event-driven timing model of the present invention;

[0081] Figure 2 This is a schematic diagram of the process for obtaining the time-series message memory of the runway operation mode according to the present invention;

[0082] Figure 3 This is a schematic diagram of the dynamic graph model for predicting runway operation modes according to the present invention. Detailed Implementation

[0083] To better understand the above-described objectives, features, and advantages of the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other. Furthermore, the present invention can be implemented in other ways different from those described herein; therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0084] A specific embodiment of the present invention, such as Figure 1-3 This invention discloses a method for predicting airport group runway operation modes based on dynamic graphs. To illustrate the effectiveness of the proposed method, a specific embodiment is provided below for detailed explanation of the above technical solution. The specific implementation steps are as follows:

[0085] Step 1: Based on historical data of the airport cluster, obtain the operation modes of multiple runways in the airport cluster and their corresponding influencing factors, and construct the node feature matrix of multiple runway operation modes;

[0086] Preferably, the specific steps for constructing the node feature matrix in step 1 include:

[0087] Using the runway operation mode i as a node, obtain the influencing factors corresponding to the runway operation mode, and establish node features v based on the influencing factors. i ;

[0088] υ i =[υ i1 ,υ i2 ,...,υ iNUM ]

[0089] Among them, υ inum Let num be the influence value of the node corresponding to runway operation mode i and the num-th influencing factor, where num = 1, 2, 3…NUM, NUM is the total number of influencing factors, and i = 1, 2, 3…N, where N represents the total number of runway operation modes.

[0090] Construct a node feature matrix for multiple runway operation modes.

[0091] Furthermore, the influencing factors corresponding to the runway operation mode include: runway operation mode usage frequency, average duration of runway operation mode, correlation between runway operation modes, weather sensitivity, flight traffic sensitivity, airport cluster effect and / or sensitivity to triggering airport rule events.

[0092] The expression for the correlation between the runway operation modes is:

[0093]

[0094] Among them, R ij u represents the correlation between runway operation mode i and runway operation mode j. ij The number of times runway operation mode i is transformed into runway operation mode j; i, j∈N, i≠j, N represents the total number of runway operation modes; T is the total time period for runway operation mode prediction.

[0095] The runway operation mode weather sensitivity measure is the strength or preference of the association between a specific runway operation mode and a specific weather condition.

[0096] The flight traffic sensitivity refers to the frequency of use of runway operation mode i under flight traffic conditions; for example, the frequency of use of runway operation mode i when flight traffic is high, and the frequency of use of runway operation mode i when flight traffic is low.

[0097] The airport cluster effect refers to the status and influence of an airport within an airport cluster. For a given airport, its effect within the airport cluster is determined by its traffic correlation and distance from other airports.

[0098] It is understood that the triggering airport rule event sensitivity refers to the frequency of use of runway operation mode i in the event of an unsafe event in the airport group; the unsafe event includes loss of control in the air, runway safety, and controlled flight into the ground.

[0099] Step 2: Treat the mode transition between the two runway operation modes as an edge, and obtain the influencing factors corresponding to the mode transition; obtain the edge features based on the influencing factors corresponding to the mode transition;

[0100] Construct a set of edge features corresponding to multiple runway operation modes;

[0101] It is understood that the mode switching between the two runway operation modes includes: based on specific conditions (e.g., weather conditions, flight traffic volume, flight safety requirements, or specific time periods), in order to ensure the efficient operation of aircraft in the airport group, the runway operation mode i is switched to the adjacent runway operation mode j.

[0102] The influencing factors corresponding to the mode transition include: the frequency of the transition between the two runway operation modes, the average transition delay, the correlation between the frequency of the transition between the two runway operation modes, the correlation between the transition and the weather, the flight traffic during the transition, the airport cluster effect of the transition and / or the airport rule events triggered by the transition.

[0103] The correlation of the switching frequency of the two runway operation modes is: the probability of switching from one runway operation mode to another and then to a third runway operation mode. Assuming an airport has three runway operation modes: A, B and C, if we observe that within a specific time period, the airport switches from runway operation mode A to runway operation mode B, and then frequently switches to runway operation mode C, then the correlation of this switching sequence A→B→C will be relatively high.

[0104] The weather sensitivity of the two runway operation modes refers to the impact of specific weather conditions on an airport's switch from one runway operation mode to another; under certain weather conditions, such as rain or fog, an airport may be more inclined to switch from the current mode to a specific, safer mode.

[0105] It is understandable that the airport cluster effect of the conversion of the various runway operation modes is based on the distance and flight traffic between airports in airport cluster g, where runway operation mode i is converted from airport a to airport b, a, b∈g, and g represents an airport cluster consisting of multiple airports.

[0106] Step 3: Obtain the feature set based on the node feature matrix and the edge feature set;

[0107] A dynamic graph model is constructed, and the feature set and the event-driven time model are embedded into the dynamic graph model to obtain a time-series dynamic graph model;

[0108] Use a time-series dynamic graph model to obtain the time-series message memory of multiple runway operation modes at the current moment;

[0109] The time-series message memory of multiple runway operation modes at the current moment is used to build a dynamic graph model for predicting the runway operation mode;

[0110] Preferably, acquiring the time-series message memory of multiple runway operation modes at the current moment specifically includes:

[0111] A dynamic graph model is constructed, and the feature set and the event-driven time model are embedded into the dynamic graph model to obtain a time-series dynamic graph model;

[0112] Determine whether node i in runway operation mode at the current time t is an independent node;

[0113] It is understandable that at the current time t, runway operation mode i is an independent node representing whether the next runway operation mode predicted by runway operation mode i is still runway operation mode i.

[0114] Runway operation mode i is not an independent node, which means that runway operation mode i has neighboring nodes. Runway operation mode i and its neighboring nodes will switch between runway operation modes.

[0115] If node i in runway operation mode at current time t is an independent node, the feature v of node i in runway operation mode at current time t is obtained based on the node feature matrix. i (t); The current time t runway operation mode i node feature v i (t) Input the time-series dynamic graph model to obtain the runway operation mode node features v before the current time t. i Global information s of (t) i (t-), t=1,2,3…T, where T represents the total number of time points;

[0116] It is understandable that the runway operation mode i node feature v before the current time t... i Global information s of (t) i (t-)′ represents the memory corresponding to all events that occur from time 0 to time t;

[0117] Using a graph attention mechanism, the features v of node i representing the runway running mode before the current time t are... i Global information s of (t) i (t-)′ and the current time t, runway operation mode i, node characteristics v i (t) is aggregated to obtain the initial aggregation message m of runway operation mode i at the current time t.i (t)′;

[0118] Obtain multiple events of runway operation mode i before the current time t, and aggregate the message m of runway operation mode i at the current time t based on the multiple events. i (t)′ performs batch aggregation; the multiple events correspond to multiple times, and the times corresponding to the multiple events are in an unordered state, where the time of the latest event is t, and there are a total of B events;

[0119] Obtain the aggregated message of runway operation mode i at the current time t.

[0120] Aggregate the messages of runway operation mode i at the current time t. Features of node i in the runway operation mode before the current time t i Global information s of (t) i (t-)′ performs message propagation and updates to obtain the current time t, runway operation mode i node features v i Global information s of (t) i (t)′;

[0121] The current time t runway operation mode node feature v i Global information s of (t) i (t)′、Runway operation mode i node characteristics v i (t) performs message propagation and generates a time-series message memory Z for the runway operation mode i of the independent node at the current time t. i(t) If node i in runway operation mode at current time t is not an independent node, obtain the node feature v in runway operation mode i at current time t based on the node feature matrix and edge feature set. i (t) and its neighboring runway operation mode j node characteristics v j (t), to obtain the edge features e of runway operation mode i and runway operation mode j at the current time t. ij (t); i≠j, i, j∈N, N represents the total number of runway operation modes;

[0122] The runway operation mode i node feature v i (t) and runway operation mode j node characteristics v j (t) Input the time-series dynamic graph model to obtain the runway operation mode i node features v before the current time t. i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) j (t-), Current time t edge feature e ij(t); Using a graph attention mechanism, the features v of node i representing the runway running mode before the current time t are... i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) i (t-), Current time t edge feature e ij (t), current time t, runway operation mode, node characteristics v i (t) is aggregated to obtain the initial aggregation message m of multiple nodes in runway operation mode i at the current time t. i (t); j∈N, N represents the total number of runway operation modes;

[0123] Obtain multiple events of runway operation mode i before time t at the current time t, and based on the multiple events before time t, generate the multi-node initial aggregation message m of runway operation mode i at the current time t. i (t) Perform batch aggregation to obtain the multi-node aggregation message of runway operation mode i at the current time t. It is understandable that the multiple events before the current time t correspond to different times, which is an unordered time state, where the latest event occurs at time t, and there are a total of B events;

[0124] Aggregate the multi-node message of runway operation mode i at the current time t. Features of node i in the runway operation mode before the current time t i Global information s of (t) i (t-), Runway operation mode j node characteristics before current time t, v j Global information s of (t) j (t-) Perform message propagation and update to obtain the current time t, runway operation mode i node features v i Global information s of (t) i (t) and the current time t, runway operation mode, node characteristics v j Global information s j (t);

[0125] The current time t runway operation mode i node feature v i Global information s of (t) i (t) and the current time t, runway operation mode, node characteristics v j Global information s of (t) j (t), current time t edge feature e ij (t), current time t, runway operation mode, node characteristics v i (t) and the current time t, runway operation mode, node characteristics v j(t) performs message propagation and generates the time-series message memory Z of the non-independent node runway operation mode i at the current time t. i(t) ;

[0126] By iterating through multiple runway operation modes and repeating the above steps, we can obtain the time sequence message memory of multiple runway operation modes at the current time t.

[0127] Furthermore, the time-series message memory Z of the non-independent node runway operation mode i at the current time t. i(t) The expression is:

[0128]

[0129] Among them, Z i(t) Let `emb(i, t)` be the temporal message memory of runway operation mode i of non-independent node at time `t`; `emb(·)` is the embedding function; `emb(i, t)` is the information collected from the neighbor runway operation modes of runway operation mode i at time `t` through the graph attention mechanism; `Attention(·)` is the graph attention weight, and `v` is the value of `i`. i Let vi be the feature of node i in runway operation mode, and vj be the feature of node j in runway operation mode, where i = 1, 2, 3…N. Let ei be the k-th order neighborhood of runway operation mode i during the process from time 0 to time t; i, j∈N, where N represents the total number of runway operation modes, k=1,2,3…K, and K represents the total order of the neighborhood; j For edge features; s j (t) represents the node characteristics of runway operation mode j at the current time t. j The global information of runway operation mode i at current time t; si(t) is the global information of runway operation mode i node vi at current time t; vi(t) is the global information of runway operation mode i node at current time t; v j (t) represents the feature of node j in runway operation mode at the current time t; h(·) is the learning function.

[0130] h(s i (t),s j (t),e ij ,υ i (t),υ j (t))=α ij ·MLP(s i (t)||s j (t)||e ij ||υ i (t)||υ j (t))

[0131] Where, α ijThe attention coefficient is calculated based on the node features of runway operation mode i and runway operation mode j; MLP(·) is a nonlinear function used to learn complex nonlinear representations from the input features.

[0132] The technical solution of this invention obtains the initial aggregated message of node i in the runway operation mode at the current time t based on the graph attention mechanism. This message is used to capture the key features of the node at a specific time and its relationship with other nodes. The initial aggregated messages of the runway operation mode at different time points are aggregated to reflect the dynamic changes of the node over time. While maintaining the time series information of the node, the characteristics of the node at different time points are further integrated. The initial aggregated messages of the runway operation mode at different time points are further aggregated in the time dimension to obtain the aggregated message, which more comprehensively reflects the dynamic characteristics of the node over time. This two-stage aggregation method helps to more accurately capture and understand the behavior of nodes in the time series, thereby providing support for time- and event-based prediction models.

[0133] This invention employs an event-driven mechanism that triggers predictions based on real-time changes in data characteristics. The core of this method lies in monitoring key data indicators and immediately performing predictive analysis when these indicators show significant changes. For example, events that trigger prediction updates include adding a new flight schedule to the system or when sudden weather changes affect runway availability. The dynamic graph model captures these changes in real time and immediately reassesses the operational patterns of the airport complex runways, generating updated predictions. This approach enables the prediction model to respond more sensitively to dynamic changes in actual operations, improving the timeliness and accuracy of predictions, avoiding unnecessary calculations when data is relatively static, thereby optimizing resource utilization and response speed.

[0134] Event-driven forecasting means that model updates are triggered based on the occurrence of events, rather than at fixed time intervals. In this case, events might be: the scheduling or cancellation of new flights; predicted weather events, such as the approach of a storm; unusual flight delays or emergencies; changes in runway approach and exit status. As these events occur, the dynamic graphical model receives new data points and then updates its forecasts based on these data points. This approach is more flexible and adaptable, providing more timely and accurate information for operational decisions.

[0135] Step 4: Based on the runway operation mode prediction dynamic graph model and loss function, predict the runway operation mode at the next time of the current time t runway operation mode i, and obtain the prediction probability of each runway operation mode as the runway operation mode at the next time of the current time t runway operation mode i, thus obtaining multiple prediction probabilities.

[0136] A threshold is set, and the multiple predicted probabilities are compared with the threshold. The runway operation mode that is greater than the threshold is taken as the prediction result of the runway operation mode at the next time step of the current time step t runway operation mode i, so as to obtain multiple prediction results corresponding to the runway operation mode at the next time step of the current time step t runway operation mode i.

[0137] Preferably, when runway operation mode i at the current time t is a non-independent node, the prediction probability expression for the runway operation mode at the next time step of the non-independent node runway operation mode i at the current time t is:

[0138]

[0139] Among them, P iq,t+1 This represents the predicted probability that the runway operation mode i of a non-independent node will be runway operation mode q at the next time t+1. Z represents the sigmoid function, i, q∈N, where N represents the total number of runway operation modes, and Z... q(t) Z is the time-series message memory of the non-independent node runway operation mode q at the current time t; i(t) This is the time-series message memory of the non-independent node runway operation mode i at the current time t, where T is the transpose. Let be the dot product vector of the non-independent node runway operation modes i and q at the current time t.

[0140] The loss function expression for runway operation mode i at the current time t, where it is a non-independent node, is as follows:

[0141]

[0142] Among them, y iq,t+1 Let y represent the link relationship between runway operation mode i and runway operation mode q at time t+1 (non-independent node). iq,t+1 This indicates that the link exists, y iq,t+1 =0 indicates that the link does not exist.

[0143] When runway operation mode i is an independent node at the current time t, the prediction probability expression for the runway operation mode at the next time step of the independent node runway operation mode i at the current time t is:

[0144]

[0145] Among them, P′ iq,t+1 This represents the predicted probability that the runway operation mode of an independent node i at the next time t+1 is runway operation mode q. Z represents the sigmoid function, i, q∈N, where N represents the total number of runway operation modes, and Z... q(t)′Z is the time-series message memory of the independent node runway operation mode q at the current time t; i(t)′ This is the time-series message memory of the independent node runway operation mode i at the current time t, where T is the transpose. Let be the dot product vector of the independent node runway operation modes i and q at the current time t.

[0146] Furthermore, thresholds are set based on historical runway usage data, flight schedule data, environmental and meteorological data, operational constraints and rules, runway condition data, node and link relationship data, and tag data.

[0147] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting runway operation modes of airport clusters based on dynamic graphs, characterized in that, include: Step 1: Based on historical data of the airport cluster, obtain the operation modes of multiple runways in the airport cluster and their corresponding influencing factors, and construct the node feature matrix of multiple runway operation modes; Step 2: Treat the mode transition between the two runway operation modes as an edge, and obtain the influencing factors corresponding to the mode transition; obtain the edge features based on the influencing factors corresponding to the mode transition; Construct a set of edge features corresponding to multiple runway operation modes; Step 3: Obtain the feature set based on the node feature matrix and the edge feature set; A dynamic graph model is constructed, and the feature set and the event-driven time model are embedded into the dynamic graph model to obtain a time-series dynamic graph model; The time-series dynamic graph model is used to obtain the time-series message memory of multiple runway operation modes at the current moment, specifically including: Based on a time-series dynamic graph model; determine the current time step. t Is the runway operating mode an independent node? If the current time t The runway operates in an independent node mode, obtaining the current time based on the node feature matrix described in the feature set. t Runway operation mode node characteristics; The current time is obtained based on the aforementioned time-series dynamic graph model. t Initial aggregation message for independent nodes; Obtain the runway operation mode at the current moment t Previous events, based on the current moment t Previous events will affect the current moment. t Independent nodes perform initial aggregation of messages in batches and multiple message propagations to generate the current time. t Sequential message memory for independent node runway operation mode; If the runway operation mode is not an independent node; obtain the current time. t The adjacent runway operation modes of the runway operation mode; based on the node feature matrix and the edge feature set in the feature set, generate the node features and edge features of two adjacent runway operation modes respectively; The current time is obtained based on the aforementioned time-series dynamic graph model. t Initial aggregation message from multiple nodes; obtain the runway operating mode at the current moment. t Previous events, based on the current moment t Previous events will affect the current moment. t Multiple nodes perform initial aggregation of messages in batches and propagate messages multiple times to generate the current time. t Time-series message memory for non-independent node runway operation mode; Traverse multiple runway operation modes to obtain the time sequence message memory of multiple runway operation modes at the current moment; Current moment t Non-independent node runway operation mode i Time sequence message memory Z i(t) The expression is: in, Z i(t) For the current moment t Non-independent node runway operation mode i Time-series message memory, t =1,2,3… T , T Indicates the total number of moments; emb ( • ) is an embedded function; emb ( i , t (This refers to the current time) t Using graph attention mechanisms to analyze runway operation patterns i Information is collected in the neighboring runway operation mode; Attention (•) represents the graph attention weights. v i Runway operation mode i Node characteristics v j For neighboring runway operation mode j Node characteristics i =1,2,3… N , From time 0 to the current time t During the process, runway operation mode i of k nearest neighbor; i , j ∈N, where N represents the total number of runway operation modes; e ij Edge features; s j ( t (This refers to the current time) t Runway operation mode j Node features v j Global information; s i ( t (This refers to the current time) t Runway operation mode i Node features v i Global information; v i ( t (This refers to the current time) t Runway operation mode i Node characteristics; v j ( t (This refers to the current time) t Runway operation mode j Node characteristics; h (•) represents the learning function; A dynamic graph model for predicting runway operation modes is established based on the time-series message memory of multiple runway operation modes at the current moment; Step 4: Use the runway operation mode prediction dynamic graph model and loss function to obtain multiple prediction results corresponding to the runway operation mode at the next time step of the current time step.

2. The airport group runway operation mode prediction method according to claim 1, characterized in that, The influencing factors corresponding to the runway operation mode described in step 1 include: runway operation mode usage frequency, average duration of runway operation mode, correlation between runway operation modes, weather sensitivity, flight traffic sensitivity, airport cluster effect and / or sensitivity to triggering airport rule events.

3. The airport group runway operation mode prediction method according to claim 2, characterized in that, The influencing factors corresponding to the mode transition relationship include: the frequency of transition between the two runway operation modes, the average transition delay, the weather correlation of the transition, the flight traffic during the transition, the airport cluster effect of the transition, the airport rule events triggered by the transition, and / or the degree of correlation between the frequency of runway operation mode transitions.

4. The method for predicting airport group runway operation modes according to claim 1, characterized in that, Step 4 uses the runway operation mode prediction dynamic graph model and loss function to obtain multiple prediction results corresponding to the runway operation mode at the next time step for the current time step. Specific steps include: Based on the runway operation mode prediction dynamic graph model and loss function, the current time... t Runway operation mode i The next runway operation mode is predicted, and the operation modes of each runway are obtained as the current time. t Runway operation mode i The predicted probability of the runway operation mode at the next moment is obtained, resulting in multiple predicted probabilities; A threshold is set, and the multiple predicted probabilities are compared with the threshold. Runway operation modes that exceed the threshold are taken as time points. t Runway operation mode i The next moment's runway operation mode prediction result is used to obtain the current moment's... t Runway operation mode i The next moment's runway operation mode corresponds to multiple prediction results.

5. The airport group runway operation mode prediction method according to claim 4, characterized in that, The current time t Runway operation mode i The probability expression for predicting the runway operating mode at the next moment is: in, P iq,t+1 Indicates the non-independent node runway operation mode i In the next moment t+ The runway operation mode of 1 is the runway operation mode. q The predicted probability; б (•)express sigmoid function, i , q ∈N, where N represents the total number of runway operation modes. Z q (t) For the current moment t Non-independent node runway operation mode q Sequential message memory; Zi(t) For the current moment t Non-independent node runway operation mode i Time-series message memory, T For transpose, For the current moment t Non-independent node runway operation mode i and q The dot product vector.

Images