A gan-based flight flow allocation scheme generation method

Abstract

The application provides a flight flow allocation scheme generation method based on a GAN, which comprises the following steps: collecting flight flow data, establishing an aerial flight network model; performing threat degree prediction to obtain a preliminary flow allocation scheme; constructing a generative adversarial network (GAN) and performing iterative training; using the trained GAN to obtain a first aerial flight flow allocation scheme; performing aerial flight situation assessment according to current flight flow data; combining a preset target to calculate a second aerial flight flow allocation scheme; and selecting a final flight flow allocation scheme from the first and second aerial flight flow allocation schemes. The application solves problems such as the need to comprehensively consider many factors in the planning and decision-making of flight network space-time configuration in a changing airport environment, and makes intelligent decisions while not being affected by time and space limitations, thereby improving the management efficiency and safety of aerial flight flow.

Description

Technical Field

[0001] This invention relates to a method for generating flight traffic allocation schemes, and more particularly to a method for generating flight traffic allocation schemes based on GANs. Background Technology

[0002] This section provides only background information relevant to this disclosure and is not necessarily prior art.

[0003] Current conventional strategies for flight traffic allocation involve first identifying bottlenecks in air traffic flow along routes, then establishing optimization models for traffic control. Traditional traffic control strategies include ground holding models, in-flight holding models, and diversion strategies. These classic methods mostly rely on expert solutions for human planning and decision-making, which is limited by time and space constraints, and the resulting plans cannot guarantee optimal strategies.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide a method for generating flight traffic allocation schemes based on GAN, which addresses the shortcomings of the existing technology.

[0006] To address the aforementioned technical problems, this invention discloses a method for generating flight traffic allocation schemes based on GANs, the method comprising the following steps:

[0007] Step 1: Collect current and historical flight traffic data and establish an air traffic network model; based on the target threat level evaluation index system and historical flight traffic data, predict the threat level, and combine the current flight traffic data and the predicted threat level to obtain a preliminary traffic allocation plan;

[0008] Step 2: Construct a Generative Adversarial Network (GAN). Use the collected current flight traffic data as the input to the generator in the GAN, and use the preliminary traffic allocation scheme as the input to the discriminator in the GAN. Iteratively train the GAN. Use the trained GAN to obtain the first air traffic allocation scheme.

[0009] Step 3: Conduct an air traffic situation assessment based on the current flight traffic data;

[0010] Step 4: Based on the results of the airborne flight situation assessment and the preset objectives, calculate the second airborne flight traffic allocation scheme;

[0011] Step 5: Select the final air traffic flow allocation scheme from the first air traffic flow allocation scheme and the second air traffic flow allocation scheme.

[0012] Furthermore, the current and historical flight traffic data mentioned in step 1 includes:

[0013] Flight origin, destination, route, number of flights, and estimated flight time.

[0014] Furthermore, the establishment of the air traffic network model mentioned in step 1, that is, calculating based on the current and historical flight traffic data to obtain the air traffic network model, is as follows:

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] in, Indicates the first One flight, Indicates the number of flights. Indicates the first Flight time of each flight;

[0021] The calculation method for route density A is as follows:

[0022]

[0023] Average flight time The calculation method is as follows:

[0024] .

[0025] Furthermore, the preliminary traffic allocation scheme obtained in step 1 specifically includes:

[0026] Step 1-1, define the target threat level evaluation indicators, as follows:

[0027]

[0028] in, For target threat level, For the first The weights of each factor indicator, For the first One factor indicator;

[0029] Steps 1-2: Divide the air flight network model into three layers: target layer, criterion layer and indicator layer. Construct a fuzzy matrix for each layer of indicators. Use the elements in the matrix to represent the relationship between the factor indicators. Then use fuzzy hierarchical analysis to obtain the weights of the factor indicators.

[0030] Steps 1-3: Use linear programming to construct an optimization model;

[0031] Steps 1-4: Based on the optimization model constructed in steps 1-3, a preliminary traffic allocation scheme is obtained, as follows:

[0032] Air traffic conditions: T = {T1, T2, T3}, where T1 represents relaxed, T2 represents moderate, and T3 represents congested.

[0033] Decision variable information: R={R1, R2, R3, R4, R5, R6}, where R1 represents sector traffic, R2 represents airport traffic, R3 represents airway traffic, R4 represents sector airspace utilization, R5 represents airway traffic saturation, and R6 represents the number of security incidents.

[0034] Allocation plan: ={ , , },in, Represents waiting in the air. Represents a change of course. This represents the following interval.

[0035] Furthermore, the Generative Adversarial Network (GAN) described in step 2 specifically includes:

[0036] A generator and a discriminator are used, where the generator learns the distribution of real data and the discriminator distinguishes between real data and generated data.

[0037] Generators and discriminators use value functions The minimax game can be represented as follows:

[0038]

[0039] in, This refers to random noise; Random noise The probability distribution it follows; This refers to the probability distribution that the real data follows; This refers to the probability that the discriminator considers the image generated by the generator to be a real image; Represents the distribution of real data Expectations Indicates the distribution of the generator input. Expectations;

[0040] The training criterion for the discriminator is to make the value function Maximizing, the training criterion for the generator is to maximize the value function. Minimize the loss function of the discriminator , means as follows:

[0041]

[0042] Loss function of generator , means as follows:

[0043]

[0044] During the training of Generative Adversarial Networks (GANs), the loss functions of the generator and discriminator are alternately optimized. That is, the generator is fixed while the discriminator is optimized, and then the discriminator is fixed while the generator is optimized. This process is repeated until Nash equilibrium is reached.

[0045] Furthermore, the method of constructing the optimization model using linear programming described in steps 1-3 is as follows:

[0046] Step 1-3-1, Design the objective function as follows:

[0047]

[0048] in, The objective function in the preliminary traffic allocation scheme includes: total delay cost; This represents an in-flight waiting strategy; This represents a flight change strategy; This represents a trailing interval strategy; , and The weights represent the importance of each strategy;

[0049] Step 1-3-2, the design constraints are as follows:

[0050] The first constraint, combined with the second... Weather influencing factors for each flight To constrain additional flights in sectors under high load conditions and reduce the risk of airspace congestion:

[0051]

[0052] The second constraint, combined with the first... Sector capacity impact factor for individual flights under severe weather conditions Constraints are imposed on the permissible sector capacity under severe weather conditions:

[0053]

[0054] The third constraint is to limit the total number of flights within a sector within a specific time period:

[0055]

[0056] in, Indicates the first The strategy for each flight This indicates the maximum sector capacity. Indicates the maximum impact of severe weather. This indicates the maximum sector flow rate.

[0057] Furthermore, the iterative training of the Generative Adversarial Network (GAN) described in step 2 specifically includes:

[0058] Step 2-1: Use the collected current flight traffic data as input to the generator in the Generative Adversarial Network (GAN) and obtain a virtual traffic allocation scheme through the generator.

[0059] Step 2-2: Convert the preliminary traffic allocation scheme into an image and use it as the input discriminator for real data; wherein, the conversion of the preliminary traffic allocation scheme into an image is taken as an RGB image, and the specific method is as follows:

[0060] The values ​​of the six decision variables R in the preliminary traffic allocation scheme are combined in pairs, and the combined variable values ​​are converted to the range of 0-255 and mapped to different color channels of the RGB image. The mapped RGB image is used to represent the preliminary traffic allocation scheme.

[0061] Steps 2-3: Simultaneously input the virtual traffic allocation scheme obtained in step 2-1 into the discriminator;

[0062] Step 2-4: Use a discriminator to distinguish between the two input schemes, that is, use the FID evaluation index to calculate the similarity between the two schemes. If the similarity is greater than the threshold, return to step 2-1 for iterative training. If the similarity is less than the threshold, the training is completed.

[0063] Furthermore, step 3, which involves conducting an airborne flight situation assessment, specifically includes:

[0064] The situation includes: aircraft position, speed, altitude, route, flight schedule, flight delays, weather conditions, and air traffic control instructions;

[0065] The evaluation metrics include: sector flow, sector flow capacity ratio, airway traffic saturation, controller workload, and the impact of severe weather.

[0066] The evaluation methods include:

[0067] Step 3-1: Construct a flight traffic network knowledge graph, which includes flight, weather, control factors and the interaction between these factors; construct a target threat evaluation index system.

[0068] Step 3-2, Flight Traffic Network Situation Prediction: Use a Long Short Time Memory (LSTM) network and a model based on the Prophet method to predict sector traffic and calculate the evaluation metrics.

[0069] Furthermore, the calculation method for obtaining the second air traffic flow allocation scheme in step 4 includes:

[0070]

[0071] in, Indicates the total delay cost; C represents the flight density within a preset time period or area; C represents the congestion level within a preset time period or area. It is an airspace constraint factor, reflecting the degree to which a specific area is unavailable or restricted; Factors influencing weather conditions; Adjustment factors indicating whether a flight is in an emergency or priority mode;

[0072] By adjusting the parameters, the total delay cost can be obtained. Lowest flight density within a preset time period or area C. Congestion level within a preset time period or area; airspace limitation factor. Influencing factors of weather conditions Adjustment factors for emergency or priority flights Based on the initial traffic allocation scheme, the control strategy is determined, resulting in the second air traffic allocation scheme, described as follows:

[0073] Flight status , indicating flight In time The states at the time include: holding state, takeoff preparation state, flight state, descent preparation state, and landing state;

[0074] Description of location information , indicating flight In time The location is indicated by latitude and longitude, sector number, or airway number;

[0075] Traffic allocation decision point The details are as follows:

[0076]

[0077] Where q is the sequence number of the flight decision point.

[0078] Furthermore, step 5, which involves selecting the final flight traffic allocation scheme, includes:

[0079] Step 5-1: Establish a hierarchical structure that displays the logical relationships between each level in the aerial flight network model;

[0080] Step 5-2: Introduce experts and use fuzzy language to evaluate the importance of factor indicators at each level to obtain the fuzzy discrimination matrix;

[0081] Step 5-3: Normalize the fuzzy discrimination matrix and calculate the weight of each factor index in the flow allocation decision.

[0082] Step 5-4: Combine the weights with the corresponding evaluation values ​​to perform fuzzy comprehensive evaluation and determine the optimal decision scheme.

[0083] Beneficial effects:

[0084] 1. This invention uses historical, readily available data as a training set to train the GAN network, thus generating realistic flight traffic allocation schemes that conform to actual conditions. Furthermore, once new data becomes available, the model can be updated by retraining the GAN to ensure its accuracy.

[0085] 2. This invention proposes a deep neural network structure comprising two networks that compete against each other, continuously improving their generation and discrimination capabilities through adversarial training, thus incorporating game theory principles. This effectively avoids problems where planning is heavily influenced by decision-makers, resulting in efficient and intelligent decision-making. Furthermore, it can simultaneously consider the complex factors involved in flight traffic allocation and generate a comprehensive flight traffic allocation scheme. In practical applications, this traffic control generation method, compared to traditional allocation methods, provides more accurate and intelligent strategy support for air traffic planning and management. Attached Figure Description

[0086] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0087] Figure 1 This is a flowchart of a GAN-based flight traffic allocation scheme generation method and system of the present invention.

[0088] Figure 2 This is an example diagram of the flight traffic target threat assessment index system according to an embodiment of the present invention.

[0089] Figure 3 This is a schematic diagram of the allocation scheme conversion according to an embodiment of the present invention.

[0090] Figure 4 This is a schematic diagram of the algorithm principle based on GAN network in an embodiment of the present invention.

[0091] Figure 5 This is an example diagram of a GAN-based network structure according to an embodiment of the present invention.

[0092] Figure 6 This is a flight situation assessment component of an embodiment of the present invention.

[0093] Figure 7 This is the knowledge graph construction process according to an embodiment of the present invention.

[0094] Figure 8 This is a visualization of the flight traffic network knowledge graph in an embodiment of the present invention.

[0095] Figure 9 This is the indicator weight calculation process of an embodiment of the present invention.

[0096] Figure 10 This is a knowledge graph-based network architecture according to an embodiment of the present invention.

[0097] Figure 11 This is a GAN applicable to the allocation scheme in this embodiment of the invention.

[0098] Figure 12 This is a simulation diagram when the threshold is set to 2 in one embodiment.

[0099] Figure 13 This is a simulation diagram when the threshold is set to 3 in one embodiment. Detailed Implementation

[0100] This invention proposes a flight traffic allocation scheme generation method based on GAN (Generative Adversarial Network). It presents a deep neural network structure comprising two networks, forming a generative model where the two networks compete against each other, continuously improving their generation and discrimination capabilities through adversarial training, thus incorporating game theory principles. This method effectively avoids issues where planning is heavily influenced by decision-makers, resulting in efficient and intelligent decision-making. Furthermore, it can simultaneously consider the complex factors involved in flight traffic allocation and generate a comprehensive flight traffic allocation scheme. In practical applications, this traffic control generation method, compared to traditional allocation methods, provides more accurate and intelligent policy support for air traffic planning and management.

[0101] The overall concept of the technical solution of this invention is as follows:

[0102] This method generates air traffic flow allocation schemes by iteratively training a GAN network, aiming to generate the optimal traffic flow allocation strategy for any input air traffic flow data. The method involves iteratively training the GAN network to generate allocation schemes for different air traffic flow scenarios. This mainly includes: acquiring data on the current air traffic situation and deriving a preliminary traffic flow allocation scheme; using the data as input to the GAN generator for iterative training; conducting an air traffic situation assessment based on the current air traffic situation and existing flight data; and designing a calculation formula to generate the optimal air traffic flow allocation scheme based on the situation assessment results and preset objectives.

[0103] The technical solution mainly includes the following steps:

[0104] Collect relevant current air traffic data and establish an air traffic network model. Use the target threat level evaluation index system to predict the situation and combine it with historical records to derive a preliminary traffic allocation plan.

[0105] The collected data is used as input to the GAN network generator, and the initial allocation scheme is used as the real data input to the discriminator for iterative training of the GAN.

[0106] An air situation assessment is conducted by combining current air traffic conditions with existing flight data.

[0107] Based on the situation assessment results and the preset objectives, a calculation formula is designed to generate the optimal air traffic flow allocation scheme.

[0108] The optimal solution is derived from the solutions generated by GAN and those generated by formula.

[0109] Example 1:

[0110] like Figure 1 As shown, a method for generating flight traffic allocation schemes based on GANs includes the following steps:

[0111] S1 collects current relevant air traffic data and establishes an air traffic network model. It then uses a target threat assessment index system to predict the situation and combines historical data to derive a preliminary traffic allocation plan.

[0112] S2 uses the collected data as input to the GAN network generator, and the initial allocation scheme as the real data input to the discriminator, to iteratively train the GAN.

[0113] S3 conducts an air situation assessment by combining current air flight conditions with existing flight data;

[0114] S4 combines the results of situation assessment with preset objectives to design calculation formulas to generate the optimal air traffic flow allocation scheme.

[0115] S5 derives the optimal solution from the solutions generated by GAN and those generated by formula.

[0116] In formulating the preliminary air traffic flow allocation plan, historical data and current air traffic information were comprehensively utilized. This included historical data integration and preprocessing, historical traffic analysis, demand forecasting, capacity assessment, and finally, preliminary traffic allocation. A linear programming approach was used to construct an optimization model for traffic allocation.

[0117]

[0118]

[0119]

[0120] in, The objective function is (e.g., total delay cost). These represent decision variables (such as the take-off and landing times or route choices for each flight). It is the relevant cost or benefit coefficient. and Various constraints (such as sector capacity and time windows) are defined. In simple terms, the process involves taking data such as flight density, airspace capacity, and weather conditions as input, aiming to minimize delay costs, and adjusting weights to change the priority of the strategy. Based on different conditions, a control strategy for flight selection is determined. The priorities are as follows: if the number of flights within a sector is close to its limit, in-flight holding is prioritized; in severe weather conditions, tailgating is prioritized; and if the sector is overloaded, rerouting is prioritized.

[0121] like Figure 2 The diagram shown is an example of the flight traffic target threat assessment index system according to an embodiment of the present invention. The target threat level of a flight traffic network refers to the assessment of the potential threats and risks faced by specific targets or elements (such as sector airspace or aircraft) within the flight traffic network. The flight traffic network includes aircraft, routes, airports, navigation facilities, communication systems, air traffic management facilities, etc., which together constitute a complex aviation system. The purpose of target threat assessment is to identify and quantify the severity of threats in order to develop effective risk management strategies to protect the security and availability of the flight traffic network.

[0122] The key operational characteristics include sector traffic, en-route traffic, airport traffic, sector airspace utilization, and en-route traffic saturation. Sector traffic refers to the number of aircraft within a sector, representing the density of aircraft in a specific airspace. En-route traffic refers to the number of aircraft on a specific airway, representing the density of aircraft on that airway. Airport traffic refers to the number of aircraft at a specific airport, representing the density of aircraft at that airport. Sector airspace utilization refers to the degree of utilization of air traffic within a specific airspace, reflecting the level of congestion in that area. En-route traffic saturation refers to the ratio between the number of aircraft carried by a specific airway and its design capacity, used to measure the degree of congestion on that airway.

[0123] Air traffic controller workload primarily utilizes the coordination load. Coordination load encompasses information sharing and coordination within the air traffic management system and with other control departments, airports, and relevant stakeholders. The impact of severe weather includes: flight delay rates, flight cancellation rates, and the number of safety incidents.

[0124] Combining the quantification methods of key operational features in the multi-dimensional spatiotemporal operational characteristic system of flight traffic, this invention introduces a flight traffic network target threat assessment index, defined as follows:

[0125]

[0126] in Threat level is an assessment of the potential threats and risks faced by a target (a sector of airspace). The weights for the features can be customized, or calculated as needed using the FAHP method (Fuzzy Analytic Hierarchy Process, reference: Bian Xiaofeng, Wang Lin. Analysis of Fatigue Risk Factors of Controllers Based on Fuzzy Analytic Hierarchy Process [J]. Science and Technology Innovation and Application, 2020, (03): 45-47.). The weight representing the sector flow feature. The weight representing the feature of airway traffic flow. The weight representing the characteristic of coordinated control load. This indicates the weight of the feature representing flight delay rate. This indicates the weight of the feature representing flight cancellation rate. The weight of the feature representing the number of security incidents. Let each be a characteristic variable, where, For sector flow, For airway traffic, To coordinate and control load, For flight delay rate, For flight cancellation rate, This represents the number of security incidents.

[0127] The weights of the aforementioned factor indicators are obtained using the fuzzy hierarchical analysis method, as follows:

[0128] Step 1: Create a target threat assessment index system, create a hierarchical structure, and adjust the initial complex assessment objects to make them hierarchical, systematic, and organized;

[0129] Step 2, Adjust the fuzzy complementary matrix Transform it into a fuzzy consistent matrix;

[0130] Step 3, hierarchical sorting. The weight values ​​to be calculated. These are the conditions that need to be met. Therefore, the weight vector... Solving this problem can be transformed into finding a least-squares problem that satisfies the following conditions:

[0131]

[0132] The constraint condition for parameter a in the formula is: , The ranking of the indicators is the result of comparing the importance of each indicator relative to the previous level.

[0133] ① Hierarchical composition. When a three-level hierarchical structure exists, the same method as above is used to calculate the weights of the previous level relative to the level before that, to obtain the result. .

[0134] ② Hierarchical overall ranking. Calculate the overall importance of the top-level target layer. Then according to The size of each indicator is used to rank the indicators, and the final result corresponding to the weight of each indicator is obtained.

[0135] like Figure 3 The diagram shown is a schematic representation of the allocation scheme transformation according to an embodiment of the present invention. Prior data processing is required, where the initial data is written into a set format:

[0136] Flight conditions in the air: T = {T1, T2, T3}

[0137] T1 represents relaxed, T2 represents moderate, and T3 represents congested.

[0138] Decision variable information R = {R1, R2, R3, R4, R5, R6}

[0139] Where R1 represents sector traffic, R2 represents airport traffic, R3 represents airway traffic, R4 represents sector airspace utilization, R5 represents airway traffic saturation, and R6 represents the number of security incidents.

[0140] Allocation scheme x = {x1, x2, x3}

[0141] Where x1 represents ground holding, x2 represents air holding on the flight path, and x3 represents rerouting strategy.

[0142] Next, the initial allocation scheme needs to be used as the real data input to the GAN network discriminator in the form of an RGB image. This is achieved by using channel separation. It is known that RGB images have three color modes: red, green, and blue. The values ​​of the six decision variables R mentioned above are combined in pairs, and the variable values ​​are converted to the range of 0-255. These values ​​are then mapped to different channels of the RGB image, such as R... 1、 The value of R2 is mapped to the red channel, R 3、 The value of R4 is mapped to the green channel, R 5、 The value of R6 is mapped to the blue channel, and finally an RGB image is used to represent different allocation schemes.

[0143] Because GAN network models can generate data in more than one format than just images, their input formats are also diverse, including not only RGB images as mentioned above, but also text and other formats. Considering the decision variables such as sector traffic, airport traffic, airway traffic saturation, and the number of security incidents mentioned earlier, these variables need to be converted into text format before being input into the GAN network for training. The general steps are as follows:

[0144] Step 1 requires collecting relevant flight traffic data and performing normalization processing to ensure that indicators of different scales and ranges can be compared and processed within the same framework. After that, it may also be necessary to process outliers to ensure the accuracy and consistency of the data.

[0145] Step 2: Obtain the relevant values ​​of the decision variables mentioned above;

[0146] Step 3: Map the numerical features to text. This requires dividing the numerical range of each dimension into several categories and defining a label for each category (the specific numerical division needs to be based on the actual data). This is represented as follows:

[0147] Sector flow: High i1, Medium i2, Low i3. n ∈(0,1), n=1,2,3, i1>i2>i3

[0148] Airport traffic: Busy J1, Normal J2, Lightly loaded J3. n ∈(0,1), n=1,2,3, j1>j2>j3

[0149] Route saturation: Severe k1, Moderate k2, Slight k3. n ∈(0,1), n=1,2,3, k1>k2>k3

[0150] Security incidents: Frequent (r1), Occasional (r2), Rare (r3). n ∈(0,1), n=1,2,3, r1>r2>r3

[0151] Step 4: Combine the text labels of each feature in a certain order to form a sequence. This sequence represents a specific air traffic condition, and therefore an allocation scheme. For example, a sequence might be "i1-j1-k2-r2", which corresponds to the allocation scheme of holding in the air along a flight path.

[0152] like Figure 4 The diagram shown is a schematic of the algorithm based on a GAN network according to an embodiment of the present invention. The training of the GAN network follows these steps:

[0153] Step 1: Use the collected values ​​of the 6 decision variables R as training samples for the generator to generate an allocation scheme;

[0154] Step 2: The preliminary traffic allocation scheme derived from the threat assessment and historical information is converted into an RGB image using a channel separation method, which serves as reference data for the discriminator in the GAN network.

[0155] Step 3: Input the allocation scheme generated by the generator and the preliminary allocation scheme into the discriminator for discrimination;

[0156] Step 4: Use the Fréchet Inception Distance (FID) metric to determine whether the model training is complete. A value closer to 1 indicates a lower similarity between the generated and real samples, requiring retraining from Step 1. A value closer to 0 indicates greater similarity, signifying successful model training.

[0157] like Figure 5 The diagram shown is an example of a GAN-based network structure according to an embodiment of the present invention. The network structure consists of a generator and a discriminator. In the generator, latent variables of shape 200×25 are sampled from a standard normal distribution and then transformed into a dimension suitable for convolution through a fully connected layer. These are then fed into a one-dimensional upsampling convolutional sub-network, where they undergo three upsampling operations to generate the final allocation scheme. In the discriminator network, the generated trajectories are input into a one-dimensional convolutional layer, and then a one-dimensional downsampling convolutional sub-network is used to reduce the resolution.

[0158] The entire GAN network uses a LeakyReLU activation function, which retains the advantage of fast computation while overcoming the disadvantage that neurons do not participate in updates when the input value is negative. Dropout mechanisms are incorporated into both upsampling and downsampling networks to limit the interaction between hidden layer nodes and prevent overfitting. Dropout works by randomly "dropping" (i.e., setting) a portion of the neuron outputs to 0 during training, thereby forcing the model to learn more generalized feature representations.

[0159] set up For the first The input of each neuron, For its corresponding weight, For bias terms, If the activation function is used, then the neuron's output in the absence of Dropout is... The calculation is as follows:

[0160]

[0161] After applying Dropout, during training, it is assumed that the probability of this neuron being selected for inactivation is... Then its output becomes:

[0162]

[0163] in It is a binary variable, taking values ​​{0, 1}, representing whether a neuron is inactivated (if...). =0, then the neuron outputs 0; if If the value is 1, then the output is normal, but it is not directly divided by during forward propagation. In practice, the output of neurons that have not been deactivated is directly multiplied. ).

[0164] Batch Normalization (BatchNorm) accelerates the training process by normalizing the input of each layer of a neural network, reducing internal covariate bias. It also has a certain regularization effect by introducing additional learnable parameters (scale). and offset The use of mini-batch processing of training samples indirectly increases the model's generalization ability, thus helping to prevent overfitting. The specific steps are as follows:

[0165] In each forward propagation process, for a mini-batch of input data, S1 first calculates the mean of the batch data on each feature dimension. and variance .

[0166] S2 uses these statistics to normalize each sample in the mini-batch, giving it zero mean and unit variance:

[0167]

[0168] in It is a sample. It is a very small constant (e.g., 1e-5) used to avoid the error of dividing by zero.

[0169] S3 In order to preserve the expressive power of the network, two learnable parameters are introduced. (Scale) and (Offset) is used to adjust the normalized data:

[0170]

[0171] In this way, the network can learn whether it needs to maintain the original scale and offset, or change them to better suit the task requirements.

[0172] Figure 6 The diagram illustrates the components of flight situation assessment in this embodiment of the invention. Situation refers to the actual operational status and dynamic changes within an air traffic network at a specific moment or over a period of time. This includes aircraft position, speed, altitude, route, flight schedule, flight delays, weather conditions, air traffic management instructions, and other relevant information. Combining current air traffic conditions and existing flight data, the assessment indicators include sector flow, sector flow capacity ratio, airway traffic saturation, controller workload, and the impact of severe weather. Flight situation assessment consists of three parts: knowledge graph, target threat level, and situation prediction.

[0173] Flight traffic network situation prediction. First, numerical calculations and predictions are performed on operational characteristic indicators such as sector traffic, sector capacity ratio, airway traffic saturation, air traffic controller workload, and the impact of severe weather. For example, sector traffic is predicted using a Long Short-Term Memory Neural Network (LSTM) and a model based on the Prophet algorithm. Next, bottleneck identification is performed. This involves centrally processing all flight plan data, using a transit time prediction model to obtain transit times for critical waypoints, and statistically calculating the predicted traffic flow values ​​for these waypoints. Finally, the predicted traffic flow values ​​are compared with capacity limits to identify flight traffic bottlenecks.

[0174] Figure 7The diagram illustrates the knowledge graph construction process according to an embodiment of the present invention. For knowledge graph construction in a flight network, the process begins with knowledge extraction. The source of knowledge is a dataset, which is then used to form an ontology-based knowledge representation. Next, newly acquired knowledge from different sources is integrated and merged into a new dataset. The quality of the merged knowledge is assessed before it is added to the knowledge base to ensure its quality. Finally, an ontology is constructed to form a hierarchical ontology library, and then the knowledge graph is generated.

[0175] Figure 8 The image shown illustrates the visualization of a flight traffic network knowledge graph according to an embodiment of the present invention. After constructing the knowledge graph, visualization operations can be performed to facilitate a more intuitive understanding and manipulation of the data. The constructed flight traffic knowledge graph is imported into the Neo4j graph database, and then visualization operations are performed, following these steps:

[0176] In the Neo4j console, enter the following commands in sequence:

[0177] ① CREATE INDEX ON:Resource(uri)

[0178] ② CALL semantics.importRDF('file: / / / D: / project / 5airport / work / knowledge graph / sector knowledge graph-v2.owl', 'RDF / XML',{})

[0179] ③ MATCH (n)

[0180] WHERE n.uri CONTAINS "<http: / / www.semanticweb.org / 86150 / ontologies / 2023 / 9 / untitled-ontology-13> "

[0181] SET n.uri = substring (n.uri, 71)

[0182] RETURN n

[0183] After configuring the node styles, the knowledge graph can be visualized.

[0184] Figure 9 The diagram illustrates the indicator weight calculation process according to an embodiment of the present invention. After defining the network threat assessment indicator for flight traffic, the most crucial step is calculating the indicator weights, which is performed as follows:

[0185] (1) Create a target threat assessment index system, create a hierarchical structure, and adjust the initial complex assessment objects to make them hierarchical, systematic and organized;

[0186] (2) Adjust the fuzzy complementary matrix Transform it into a fuzzy consistent matrix;

[0187] (3) Hierarchical sorting. The weight values ​​to be calculated. These are the conditions that need to be met. Therefore, the weight vector... Solving this problem can be transformed into finding a least-squares problem that satisfies the following conditions:

[0188]

[0189] The constraint condition for parameter a in the formula is: , The ranking of the indicators is the result of comparing the importance of each indicator relative to the previous level.

[0190] Figure 10 The diagram illustrates a knowledge graph-based network architecture according to an embodiment of the present invention. FAHP is a method for multi-criteria decision-making that can help assess the threat level of different objectives. In FAHP, a hierarchical structure is first established to reflect the relationships between different levels. Then, experts provide comparative data to determine the relative importance of each level. Finally, weights are calculated and evaluated based on this data.

[0191] The general steps for determining the optimal allocation scheme are as follows:

[0192] Step 1 decomposes the flight traffic allocation decision problem into a hierarchical structure (objective, criteria, indicators). Each level has corresponding factors; the indicator level includes sector traffic, sector capacity ratio, airway traffic saturation, controller workload, and the impact of severe weather. For each pair of factors in each level, their relative importance is described using fuzzy language, and a fuzzy discrimination matrix is ​​constructed. The specific process for this step is as follows:

[0193] S1 explicitly defines the quantification criteria for fuzzy linguistic variables, converting linguistic descriptions such as "very important," "important," "average," "not very important," and "not important" into mathematical fuzzy sets. Typically, these linguistic variables are mapped to membership functions with upper and lower bounds. Very important: [0.8, 1]; Important: [0.6, 0.8]; Average: [0.4, 0.6]; Not very important: [0.2, 0.4]; Not important: [0, 0.2]. Here, each interval represents a fuzzy set, expressing the range of uncertainty in the decision-maker's subjective evaluation of the relative importance of a factor.

[0194] Taking the indicator layer as an example, S2 requires evaluation of five factors. A 5×5 fuzzy discriminant matrix needs to be constructed, where each element [i, j] represents the relative importance of the i-th factor relative to the j-th factor. This process involves: first, for each pair of factors (i, j), the decision-maker selects a fuzzy linguistic variable based on their judgment; second, according to the previously defined fuzzy linguistic scale, the selected linguistic variable is converted into a corresponding membership interval and filled into the corresponding position in the matrix; finally, this process continues until the entire fuzzy discriminant matrix is ​​filled.

[0195] Step 2: Normalize each fuzzy judgment matrix to transform it into a consistency matrix. Calculate the weight of each factor using the fuzzy judgment matrix; this can be achieved through techniques such as the fuzzy eigenvector method.

[0196] Step 3: Check if the fuzzy discrimination is consistent. If inconsistent, it needs to be adjusted. Combine the weights of each sub-objective with their evaluation values ​​to obtain a fuzzy comprehensive evaluation of each decision scheme.

[0197] Step 4: Rank the decision-making options based on fuzzy comprehensive evaluation to determine the optimal option.

[0198] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0199] Figure 11 The diagram illustrates the GAN classification applicable to the allocation scheme in this embodiment of the invention. Since the emergence of Generative Adversarial Networks (GANs), they have been rapidly evolving and iterating, resulting in a wide variety of GAN types to date. For this invention, in addition to WGAN-GP used previously, cGAN and MM-GAN can also be applied.

[0200] The method for generating flight traffic allocation schemes based on Conditional Generative Adversarial Networks (cGANs) is as follows:

[0201] S1 integrates historical flight data, including past flight records, meteorological data, airspace restriction information, and airport capacity, with real-time contextual conditions such as time and weather conditions.

[0202] S2 designs a generator and discriminator architecture based on a deep neural network and including conditional input.

[0203] S3 is trained jointly by the discriminator and the generator, and the conditional loss term is added and iteratively trained and adjusted.

[0204] S4 uses domain expert knowledge and objective indicators (such as flight efficiency, delay rate, and airspace utilization) to evaluate the effectiveness and feasibility of the generated schemes, and compares the performance of the schemes generated by cGAN with existing methods or real data in a simulated environment.

[0205] The method for generating flight traffic allocation schemes based on multimodal generative adversarial networks (MM-GAN) is as follows:

[0206] S1 collects historical flight data, including flight path, time, altitude, speed, weather conditions, and air traffic control information, and performs data preprocessing.

[0207] The S2 design model architecture includes multiple generators and one discriminator. Each generator is responsible for generating flight traffic allocation schemes under specific contextual conditions such as different time periods. The discriminator not only determines whether the generated schemes are realistic, but also distinguishes between generated schemes under different conditions, ensuring that the schemes output by each generator are not only of high quality, but also conform to their corresponding condition settings.

[0208] S3 performs iterative training and optimization of the discriminator followed by the generator;

[0209] S4 uses domain knowledge and actual operational metrics to evaluate the generated flight traffic allocation scheme. Based on the evaluation feedback, it adjusts the model parameters and conducts multiple rounds of iterative optimization until a satisfactory performance standard is achieved.

[0210] Example 2:

[0211] In one specific embodiment, a flow control simulation was performed on a sector of a city in East China. This simulation was based on a flight flow simulation system.

[0212] The principle of flow control is as follows: First, a route map is constructed using waypoint and route data. An adjacency matrix is ​​used to represent the distances between waypoints, and the distance calculation employs the spherical distance formula. Next, the time and the number of aircraft in a sector are checked. If the number of aircraft in a sector reaches a threshold, the weights of the edges passing through that sector in the route map are modified to infinity, thus blocking the path. Then, Dijkstra's algorithm is used to recalculate the shortest path on the updated route map, ensuring that newly created aircraft routes bypass overloaded sectors. After path calculation is complete, the modified edge weights in the route map are restored to maintain the accuracy of subsequent path planning. The aircraft then automatically flies according to the new route.

[0213] During the simulation, the spatial bottleneck threshold was set to 2, and the simulation system was run. When the spatial domain reached the preset bottleneck, 10 different time points were selected to observe whether the text output area contained the correct prompt information.

[0214] like Figure 12 As shown, when both aircraft AC01 and aircraft AC02 are located in sector 11 of the city, the text output area indicates that the sector has reached the aircraft number threshold (2) and displays the current number of aircraft. Then, the control strategy is operated.

[0215] like Figure 13 As shown, after implementing the control strategy and setting the bottleneck threshold to 3, the number of aircraft in Shanghai sector 11 will not exceed the preset threshold. Therefore, the newly created aircraft AC05 will fly to the right along the preset path to its destination airport. The purple path in the figure represents the current route of aircraft AC05.

[0216] After modifying the bottleneck threshold to 2, when aircraft AC05 was created, the bottleneck threshold had been reached because the number of aircraft in sector 11 of the city was 2. Therefore, the flight path of aircraft AC05 was replanned. As shown in the figure above, the replanned route first flies to the left and then to the destination, Hongqiao Airport. This simulation shows that the control strategy is effective.

[0217] In its specific implementation, this application provides a computer storage medium and a corresponding data processing unit. The computer storage medium is capable of storing a computer program, which, when executed by the data processing unit, can run the invention's content regarding a GAN-based flight traffic allocation scheme generation method, as well as some or all of the steps in various embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0218] Those skilled in the art will clearly understand that the technical solutions in the embodiments of the present invention can be implemented using computer programs and their corresponding general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of computer programs, i.e., software products. These computer program software products can be stored in a storage medium and include several instructions to cause a device containing a data processing unit (which may be a personal computer, server, microcontroller, MCU, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0219] This invention provides a method and approach for generating flight traffic allocation schemes based on GANs. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for generating a GAN-based flight flow allocation scheme, characterized in that, The method includes the following steps: Step 1: Collect current and historical flight traffic data and establish an air traffic network model; based on the target threat level evaluation index system and historical flight traffic data, predict the threat level, and combine the current flight traffic data and the predicted threat level to obtain a preliminary traffic allocation plan; Step 2: Construct a Generative Adversarial Network (GAN). Use the collected current flight traffic data as the input to the generator in the GAN, and use the preliminary traffic allocation scheme as the input to the discriminator in the GAN. Iteratively train the GAN. Use the trained GAN to obtain the first air traffic allocation scheme. Step 3: Conduct an air traffic situation assessment based on the current flight traffic data; Step 4: Based on the results of the airborne flight situation assessment and the preset objectives, calculate the second airborne flight traffic allocation scheme; Step 5: Select the final air traffic flow allocation scheme from the first air traffic flow allocation scheme and the second air traffic flow allocation scheme; Specifically, step 3, which involves conducting an airborne flight situation assessment, includes: The situation includes: aircraft position, speed, altitude, route, flight schedule, flight delays, weather conditions, and air traffic control instructions; The evaluation metrics include: sector flow, sector flow capacity ratio, airway traffic saturation, controller workload, and the impact of severe weather. The evaluation methods include: Step 3-1: Construct a flight traffic network knowledge graph, which includes flight, weather, control factors and the interaction between these factors; construct a target threat evaluation index system. Step 3-2, Flight Traffic Network Situation Prediction: Use a Long Short Time Memory (LSTM) network and a model based on the Prophet method to predict sector traffic and calculate the evaluation metrics. The calculation method described in step 4 to obtain the second air traffic flow allocation scheme includes: ； wherein, represents the total delay cost; is the flight density within a preset time period or region; C represents the congestion degree within a preset time period or region; is an airspace restriction factor, reflecting the degree of unavailability or restriction of a specific region; represents an impact factor of weather conditions; represents an adjustment factor of emergency or priority flights; By adjusting the parameters, the total delay cost can be obtained. Lowest flight density within a preset time period or area C. Congestion level within a preset time period or area; airspace limitation factor. Influencing factors of weather conditions Adjustment factors for emergency or priority flights Based on the initial traffic allocation scheme, the control strategy is determined, resulting in the second air traffic allocation scheme, described as follows: Flight status , indicating flight In time The states at the time include: holding state, takeoff preparation state, flight state, descent preparation state, and landing state; Description of location information , indicating flight In time The location is indicated by latitude and longitude, sector number, or airway number; Traffic allocation decision point The details are as follows: ； Where q is the sequence number of the flight decision point.

2. The method for generating a flight traffic allocation scheme based on GAN according to claim 1, characterized in that, The current and historical flight traffic data mentioned in step 1 include: Flight origin, destination, route, number of flights, and estimated flight time.

3. The method for generating a flight traffic allocation scheme based on GAN according to claim 2, characterized in that, The establishment of the air traffic network model mentioned in step 1 involves calculating the air traffic network model based on the current and historical flight traffic data, as detailed below: ； in, Indicates the first One flight, Indicates the number of flights. Indicates the first Flight time of each flight; The calculation method for route density A is as follows: ； Average flight time The calculation method is as follows: 。 4. The method for generating a flight traffic allocation scheme based on GAN according to claim 3, characterized in that, The preliminary traffic allocation scheme described in step 1 specifically includes: Step 1-1, define the target threat level evaluation indicators, as follows: ； in, For target threat level, For the first The weights of each factor indicator, For the first One factor indicator; Steps 1-2: Divide the air flight network model into three layers: target layer, criterion layer and indicator layer. Construct a fuzzy matrix for each layer of indicators. Use the elements in the matrix to represent the relationship between the factor indicators. Then use fuzzy hierarchical analysis to obtain the weights of the factor indicators. Steps 1-3: Use linear programming to construct an optimization model; Steps 1-4: Based on the optimization model constructed in steps 1-3, a preliminary traffic allocation scheme is obtained, as follows: Air traffic conditions: T = {T1, T2, T3}, where T1 represents relaxed, T2 represents moderate, and T3 represents congested. Decision variable information: R={R1, R2, R3, R4, R5, R6}, where R1 represents sector traffic, R2 represents airport traffic, R3 represents airway traffic, R4 represents sector airspace utilization, R5 represents airway traffic saturation, and R6 represents the number of security incidents. Allocation plan: ={ , , },in, Represents waiting in the air. Represents a change of course. This represents the following interval.

5. The method for generating a flight traffic allocation scheme based on GAN according to claim 4, characterized in that, The Generative Adversarial Network (GAN) mentioned in step 2 specifically includes: A generator and a discriminator are used, where the generator learns the distribution of real data and the discriminator distinguishes between real data and generated data. Generators and discriminators use value functions The minimax game can be represented as follows: ； in, This refers to random noise; Random noise The probability distribution it follows; This refers to the probability distribution that the real data follows; This refers to the probability that the discriminator considers the image generated by the generator to be a real image; Represents the distribution of real data Expectations Indicates the distribution of the generator input. Expectations; The training criterion for the discriminator is to make the value function Maximizing, the training criterion for the generator is to maximize the value function. Minimize the loss function of the discriminator , means as follows: ； Loss function of generator , means as follows: ； During the training of Generative Adversarial Networks (GANs), the loss functions of the generator and discriminator are alternately optimized. That is, the generator is fixed while the discriminator is optimized, and then the discriminator is fixed while the generator is optimized. This process is repeated until Nash equilibrium is reached.

6. The method for generating a flight traffic allocation scheme based on GAN according to claim 5, characterized in that, The optimization model constructed using linear programming as described in steps 1-3 is as follows: Step 1-3-1, Design the objective function as follows: ； in, The objective function in the preliminary traffic allocation scheme includes: total delay cost; This represents an in-flight waiting strategy; This represents a flight change strategy; This represents a trailing interval strategy; , and The weights represent the importance of each strategy; Step 1-3-2, the design constraints are as follows: The first constraint, combined with the second... Weather influencing factors for each flight To constrain additional flights in sectors under high load conditions and reduce the risk of airspace congestion: ； The second constraint, combined with the first... Sector capacity impact factor for individual flights under severe weather conditions Constraints are imposed on the permissible sector capacity under severe weather conditions: ； The third constraint is to limit the total number of flights within a sector within a specific time period: ； in, Indicates the first The strategy for each flight This indicates the maximum sector capacity. Indicates the maximum impact of severe weather. This indicates the maximum sector flow rate.

7. The method for generating a flight traffic allocation scheme based on GAN according to claim 6, characterized in that, Step 2, which involves iteratively training the Generative Adversarial Network (GAN), specifically includes: Step 2-1: Use the collected current flight traffic data as input to the generator in the Generative Adversarial Network (GAN) and obtain a virtual traffic allocation scheme through the generator. Step 2-2: Convert the preliminary traffic allocation scheme into an image and use it as input to the discriminator as real data; wherein, the specific method for converting the preliminary traffic allocation scheme into an image is as follows: The values ​​of the six decision variables R in the preliminary traffic allocation scheme are combined in pairs, and the combined variable values ​​are converted to the range of 0-255 and mapped onto different color channels of the image. The mapped image is then used to represent the preliminary traffic allocation scheme. Steps 2-3: Simultaneously input the virtual traffic allocation scheme obtained in step 2-1 into the discriminator; Step 2-4: Use a discriminator to distinguish between the two input schemes, that is, use the FID evaluation index to calculate the similarity between the two schemes. If the similarity is greater than the threshold, return to step 2-1 for iterative training. If the similarity is less than the threshold, the training is completed.

8. The method for generating a flight traffic allocation scheme based on GAN according to claim 7, characterized in that, Step 5 describes the selection process to obtain the final flight traffic allocation scheme, which includes: Step 5-1: Establish a hierarchical structure that displays the logical relationships between each level in the aerial flight network model; Step 5-2: Introduce experts and use fuzzy language to evaluate the importance of factor indicators at each level to obtain the fuzzy discrimination matrix; Step 5-3: Normalize the fuzzy discrimination matrix and calculate the weight of each factor index in the flow allocation decision. Step 5-4: Combine the weights with the corresponding evaluation values ​​to perform fuzzy comprehensive evaluation and determine the optimal decision scheme.

Images