Flight trajectory missing point filling method and device based on generative adversarial network
By learning the complex motion mode of high-maneuverable aerial targets based on the method of generating adversarial networks, trajectory data that conforms to the actual motion laws are generated, and the problem of large gap between the prediction results and the real trajectory in the prior art is solved, and the efficient and highly adaptable trajectory filling effect is achieved.
Patent Information
- Application Number
- CN202411882241.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-11
AI Technical Summary
In the prior art, in the filling of flight trajectory, the trajectory generated by the merger of prediction results has a large gap between the real trajectory and the situation where the trajectory missing values account for a large or relatively discrete amount.
Using a method based on generative adversarial network, a generative adversarial network model is constructed, including generators and discriminators, to learn complex motion patterns of highly maneuverable aerial targets, generate trajectory data that conforms to actual motion laws, and fill in the missing points in the flight trajectory.
It improves the integrity and availability of trajectory data, is suitable for trajectory prediction and filling of high-mobility targets, can work stably in complex environments, provide real-time or near-real-time filling results, and enhances the monitoring and prediction capabilities of high-mobility aerial targets.
Smart Images
Figure CN119807573A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for filling missing points of a flight trajectory based on a generative adversarial network. Background Art
[0002] The trajectory analysis and research of highly maneuverable aerial targets are of vital importance to the development of aircraft. The maneuverability, flexibility and high-speed movement characteristics of such targets make the collection of trajectory data face major challenges. In the process of data collection, a large number of missing data points often appear, and these missing data points pose a significant obstacle to further analysis and research. Therefore, the research on flight trajectory filling methods is particularly important. It can not only improve the overall performance and safety of aerial targets and reduce operating costs, but also enhance the ability of aircraft to cope with complex environments and meet increasingly stringent regulatory requirements. When it comes to filling the flight trajectories of highly maneuverable aerial targets, there is no dedicated and efficient method to solve this problem. At present, trajectory filling can be completed by converting the filling problem into a prediction problem:
[0003] Physical model-based methods: Advantages: Use known physical laws, such as Newton's laws of motion, to predict the trajectory of the aircraft. This method has good interpretability and is easy to understand the physical reasons for the trajectory changes. It has high accuracy in theory, especially in the short term. Disadvantages: The model may be too complex to handle all the variables encountered in actual flight, such as wind speed changes, aircraft performance changes, etc.
[0004] Data-driven methods: Advantages: Able to learn the flight mode of aircraft from historical data, such as the combination of long short-term memory network and differential autoregressive moving average model, which can handle nonlinear and non-stationary time series and improve the accuracy of prediction. Disadvantages: A large amount of historical data is required to train the model, and the generalization ability of the model is limited by the quality and diversity of the training data.
[0005] Methods based on Monte Carlo simulation: Advantages: Generate a large number of possible trajectories by random sampling, which can provide the probability distribution of trajectories and help assess uncertainty and risk. Disadvantages: The computation is large, and a large number of samples are required to ensure the accuracy of the results. The accuracy of the results is limited by the number and quality of samples.
[0006] Methods based on hybrid models: Advantages: It combines the advantages of physical models and data-driven models. By using algorithms such as Kalman filtering and particle filtering, it can improve the accuracy and robustness of predictions. Disadvantages: The construction and maintenance of the model are relatively complex, and professional knowledge is required to adjust and optimize the model parameters.
[0007] There are two problems when the trajectory filling problem is transformed into multiple prediction problems. First, the trajectory generated by merging multiple prediction results is quite different from the real trajectory. Second, this method is more suitable for trajectories with a small or concentrated proportion of missing values, and it is more difficult to solve trajectories with a large or discrete proportion of missing values. Summary of the invention
[0008] In order to solve the problems in the prior art of flight trajectory filling that the trajectory generated by merging the prediction results is quite different from the actual trajectory and that it is difficult to solve the problem that the missing values of the trajectory account for a large proportion or are relatively discrete, the present invention proposes a flight trajectory filling method and device based on a generative adversarial network. By designing a generative adversarial network structure, it is possible to learn the complex motion patterns of highly maneuverable aerial targets and generate trajectory data that conforms to the actual motion laws.
[0009] In a first aspect, the present invention provides a method for filling missing points in a flight trajectory based on a generative adversarial network, comprising:
[0010] Step 1: Collect the flight trajectory data of the aircraft and preprocess it to obtain a flight trajectory matrix set;
[0011] Step 2: construct a generative adversarial network model; wherein the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data;
[0012] Step 3: training the adversarial network model based on the flight trajectory matrix set to obtain a flight trajectory filling model;
[0013] Step 4: The flight trajectory data to be filled is pre-processed and then input into the flight trajectory filling model to generate filled flight trajectory data.
[0014] Furthermore, the step 1 specifically includes:
[0015] Step 1.1: Collect the flight trajectory data of the aircraft and extract the data of six attributes: longitude, latitude, altitude, horizontal speed, angle and vertical speed;
[0016] Step 1.2: Based on the horizontal speed and the angle, the horizontal speed is decomposed into a speed Vx along the longitude direction and a speed Vy along the latitude direction according to the trigonometric function theorem, and the formula is as follows:
[0017] Vx=V*cos(Head)
[0018] Vy=V*sin(Head)
[0019] Among them, V represents the horizontal speed, and Head represents the angle;
[0020] Step 1.3: Using a difference calculation method, the longitude and the latitude are converted into a longitude difference and a latitude difference, and the formula is as follows:
[0021] Lon_Diff=|Lon j+1 -Lon j |*100000
[0022] Lat_Diff=|Lat j+1 -Lat j |*100000
[0023] Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j represents the latitude at time j;
[0024] Step 1.4: Set the minimum altitude value to N, calculate the difference between the lowest altitude in the flight trajectory data and N, and subtract the difference from the altitude to get the relative altitude Alt:
[0025] Alt=Alt_Real-(Alt_Real(Min)-N)
[0026] Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data;
[0027] Step 1.5: intercepting the flight trajectory data according to the timestamp, wherein the intercepted data of each sequence includes the longitude difference, the latitude difference, the absolute altitude, the speed in the longitude direction, the speed in the latitude direction and the vertical speed;
[0028] Step 1.6: The intercepted sequence is made into a flight trajectory matrix according to the missing data, wherein the position of the missing data in the flight trajectory matrix is represented by 0, and the position without missing data is represented by 1.
[0029] Furthermore, the generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module and a fully connected module; the word embedding module includes a one-dimensional convolutional neural network layer; the downsampling module includes a one-dimensional convolutional neural network layer; the Transformer module includes a three-layer Transformer encoder; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer and a skip connection layer; the fully connected module includes a three-layer fully connected neural network layer.
[0030] Furthermore, the discriminator includes a one-dimensional convolutional network, a global pooling layer and a fully connected layer.
[0031] Furthermore, in step 3, a binary cross entropy loss function is used to calculate the loss value of the discriminator output value, a mean square error loss function is used to calculate the loss value of the generator output value, and an Adam optimizer is used to iteratively train the generator and the discriminator.
[0032] In a second aspect, the present invention provides a flight trajectory missing point filling device based on a generative adversarial network, comprising:
[0033] A data acquisition preprocessing module is used to collect and preprocess the flight trajectory data of the aircraft to obtain a flight trajectory matrix set;
[0034] A generative adversarial network model module, used to construct a generative adversarial network model; wherein the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data;
[0035] A flight trajectory filling model module, used for training the adversarial network model based on the flight trajectory matrix set to obtain a flight trajectory filling model;
[0036] The data output module is used to input the flight trajectory data to be filled into the flight trajectory filling model after preprocessing, so as to generate the filled flight trajectory data.
[0037] Furthermore, the data acquisition preprocessing module is specifically used for:
[0038] Collect the flight trajectory data of the aircraft and extract the data of six attributes: longitude, latitude, altitude, horizontal speed, angle and vertical speed;
[0039] Based on the horizontal speed and the angle, the horizontal speed is decomposed into a speed Vx along the longitude direction and a speed Vy along the latitude direction according to the trigonometric function theorem, and the formula is as follows:
[0040] Vx=V*cos(Head)
[0041] Vy=V*sin(Head)
[0042] Among them, V represents the horizontal speed, and Head represents the angle;
[0043] The longitude and the latitude are converted into longitude difference and latitude difference by using the difference calculation method, and the formula is as follows:
[0044] Lon_Diff=|Lon j+1 -Lon j |*100000
[0045] Lat_Diff=|Lat j+1 -Lat j |*100000
[0046] Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j represents the latitude at time j;
[0047] Set the minimum altitude value to N, calculate the difference between the lowest altitude in the flight trajectory data and N, and subtract the difference from the altitude to get the relative altitude Alt:
[0048] Alt=Alt_Real-(Alt_Real(Min)-N)
[0049] Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data;
[0050] intercepting the flight trajectory data according to timestamps, wherein each sequence of intercepted data includes the longitude difference, the latitude difference, the absolute altitude, the speed in the longitude direction, the speed in the latitude direction and the vertical speed;
[0051] The intercepted sequence is made into a flight trajectory matrix according to the missing data, wherein the position of the missing data in the flight trajectory matrix is represented by 0, and the position without missing data is represented by 1.
[0052] Furthermore, the generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module and a fully connected module; the word embedding module includes a one-dimensional convolutional neural network layer; the downsampling module includes a one-dimensional convolutional neural network layer; the Transformer module includes a three-layer Transformer encoder; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer and a skip connection layer; the fully connected module includes a three-layer fully connected neural network layer.
[0053] Furthermore, the discriminator includes a one-dimensional convolutional network, a global pooling layer and a fully connected layer.
[0054] Furthermore, in the flight trajectory filling model module, a binary cross entropy loss function is used to calculate the loss value of the discriminator output value, a mean square difference loss function is used to calculate the loss value of the generator output value, and an Adam optimizer is used to iteratively train the generator and the discriminator.
[0055] Beneficial effects of the present invention:
[0056] (1) Based on the powerful generative capability of generative adversarial networks, the present invention can generate missing data that is consistent with the distribution of original trajectory data, thereby improving the integrity and availability of trajectory data. This method is particularly suitable for predicting and filling the trajectory of highly maneuverable targets, such as fighter jets or high-speed drones, and can continue to provide possible flight paths when the target is out of observation. It has important practical application value in the fields of military defense, air traffic management, and target tracking.
[0057] (2) The present invention has strong adaptability and robustness. Whether in a complex electromagnetic environment or facing the sudden maneuver of the target, the method and device of this patent can work stably, and can effectively handle different types of highly maneuverable aerial targets and different degrees of track loss, providing reliable technical support for the monitoring and tracking of aerial targets.
[0058] (3) The present invention can quickly process large amounts of data and provide real-time or near real-time filling results, thereby enhancing the monitoring and prediction capabilities of highly maneuverable aerial targets.
[0059] (4) The present invention realizes an automated and intelligent processing flow, reduces human intervention, and improves the speed and accuracy of data processing, which is of great significance for improving the technical level in the fields of air traffic monitoring, military target tracking and air safety.
[0060] (5) Through the carefully designed generative adversarial network structure, the present invention is able to learn the complex motion patterns of highly maneuverable aerial targets and generate trajectory data that conforms to the actual motion laws, effectively filling in the missing trajectories caused by the limitations of observation equipment or the maneuverability of the target. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 A schematic diagram of a flow chart of a flight trajectory filling method based on a generative adversarial network provided in an embodiment of the present invention;
[0062] Figure 2 A schematic diagram of the structure of a generative adversarial network provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0064] like Figure 1 As shown, an embodiment of the present invention provides a method for filling missing points of a flight trajectory based on a generative adversarial network, comprising:
[0065] Step 1: Collect the flight trajectory data of the aircraft and preprocess it to obtain a flight trajectory matrix set;
[0066] Specifically, step 1.1: collect the flight trajectory data of the aircraft and extract the data of six attributes: longitude Lon, latitude Lat, altitude Alt_Real, horizontal velocity V, angle Head and vertical velocity Vertical Velocity;
[0067] Step 1.2: Process the velocity data. Based on the data of horizontal velocity and angle, the horizontal velocity is decomposed into the velocity Vx along the longitude direction and the velocity Vy along the latitude direction according to the trigonometric function theorem. The formula is as follows:
[0068] Vx=V*cos(Head)
[0069] Vy=V*sin(Head)
[0070] Among them, V represents the horizontal speed, and Head represents the angle;
[0071] Step 1.3: Process the longitude and latitude data. In order to improve the accuracy of the filling results, the longitude and longitude data are processed. The difference calculation method is used to convert the longitude and latitude into the longitude difference and latitude difference. This can effectively smooth the mutation in the data and improve the performance. The formula is as follows:
[0072] Lon_Diff=|Lon j+1 -Lon j |*100000
[0073] Lat_Diff=|Lat j+1 -Lat j |*100000
[0074] Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j Represents the latitude at time j.
[0075] Step 1.4: Process the height data, convert the absolute distance of the height into a relative distance, set the minimum height value to N, calculate the difference between the actual minimum height and N, and subtract the difference from each height data to get the relative height. The conversion formula is as follows:
[0076] Alt=Alt_Real-(Alt_Real(Min)-N)
[0077] Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data.
[0078] Step 1.5: Intercept the data. Intercept the flight trajectory data according to the timestamp. Each intercepted sequence of data includes six attributes: longitude difference, latitude difference, absolute altitude, speed along longitude, speed along latitude, and vertical speed.
[0079] It can be understood that the format of the input data is [60, 6], and the embodiment of the present invention is used to fill in the position of the flight trajectory, including longitude, latitude, and altitude. The speed data belongs to auxiliary data, and the speed along the longitude direction, the speed along the latitude direction, and the vertical speed are filled in by using a sliding window average method, so that it can better assist in filling in the trajectory.
[0080] Step 1.6: Matrix preparation: The intercepted sequence is prepared into a flight trajectory matrix based on the missing data, where the position of the missing data in the flight trajectory matrix is represented by 0, and the position without missing data is represented by 1.
[0081] Specifically, the truncated sequence is made into a 0-1 matrix, as shown in Table 1 and Table 2. The 0-1 matrix is used to represent the position of the missing value, where the position of the missing value is represented by 0, and the position without missing value is represented by 1.
[0082] Table 1: Trajectory data with missing values
[0083] 1 / 3 41 5 / 1 2 3 / / 6 / 2 / / / 6
[0084] Table 2: Matrix M indicating the location of missing values
[0085] 1 0 1 1 1 0 1 1 1 0 0 1 0 1 0 0 0 1
[0086] Step 2: Construct a generative adversarial network model; the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data.
[0087] Specifically, a U-Net is constructed as a generator based on Transformer and one-dimensional convolutional neural network to reconstruct the entire trajectory segment, and finally the non-missing values extracted from the original data and the filled positions of the reconstructed trajectory are merged to generate the final result. The generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module and a fully connected module, wherein the word embedding module includes a one-dimensional convolutional neural network, which is responsible for word embedding encoding of the input data; the downsampling module includes a one-dimensional convolutional neural network, which is responsible for reducing the dimension of the data after embedding encoding to reduce the adverse effects of irrelevant features on the final result; the Transformer module includes a three-layer Transformer encoder, which is used to extract features from the output of the downsampling module and use its features as input data of the upsampling module through bridging; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer and a jump connection layer, which are used to reconstruct the trajectory data; the fully connected module includes three layers of fully connected neural network layers, which are used to output the final reconstructed trajectory.
[0088] The discriminator consists of a one-dimensional convolutional network, a global pooling layer, and a fully connected layer.
[0089] Specifically, the discriminator includes three layers of one-dimensional convolutional neural network layers, a global pooling layer and a fully connected neural network layer; the discriminator first extracts data features through three layers of one-dimensional convolutional neural network, and then uses the one-dimensional convolutional neural network layer to perform jump connections on it, and then uses the global average pooling layer to perform global average pooling on each feature map, and finally classifies the input data through three layers of fully connected layers to determine whether the data is real data.
[0090] Step 3: Train the adversarial network model based on the flight trajectory matrix set to obtain the flight trajectory filling model;
[0091] Specifically, Figure 2 As shown, the mean square error loss function is used to calculate the loss value Loss_G for the generator, and the binary cross loss function is used to calculate the loss value Loss_D for the discriminator. The Adam optimizer is used to iteratively train the discriminator and generator. The discriminator uses the calculated Loss_D for back propagation, and the generator uses the calculated (Loss_D+Loss_G) / 2 for back propagation.
[0092] It can be understood that a new type of generator and a new type of discriminator are proposed to form a generative adversarial network model. The generator is responsible for filling the missing points of the trajectory, and the discriminator is responsible for judging whether the sequence filled by the generator is a true sequence. The generator and the discriminator grow together in the confrontation, and finally reach a balance and output the final result.
[0093] Step 4: The flight trajectory data to be filled is pre-processed and then input into the flight trajectory filling model to generate filled flight trajectory data.
[0094] The embodiment of the present invention further provides a flight trajectory missing point filling device based on a generative adversarial network, comprising:
[0095] A data acquisition preprocessing module is used to collect and preprocess the flight trajectory data of the aircraft to obtain a flight trajectory matrix set;
[0096] Specifically used for: collecting the flight trajectory data of the aircraft and extracting data of six attributes: longitude, latitude, altitude, horizontal speed, angle and vertical speed;
[0097] Based on the horizontal speed and angle, the horizontal speed is decomposed into the speed Vx along the longitude direction and the speed Vy along the latitude direction according to the trigonometric function theorem. The formula is as follows:
[0098] Vx=V*cos(Head)
[0099] Vy=V*sin(Head)
[0100] Among them, V represents the horizontal speed, and Head represents the angle;
[0101] The difference calculation method is used to convert longitude and latitude into longitude difference and latitude difference. The formula is as follows:
[0102] Lon_Diff=|Lon j+1 -Lon j |*100000
[0103] Lat_Diff=|Lat j+1 -Lat j |*100000
[0104] Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j represents the latitude at time j;
[0105] Set the minimum altitude value to N, calculate the difference between the lowest altitude in the flight trajectory data and N, and subtract the difference from the altitude to get the relative altitude Alt:
[0106] Alt=Alt_Real-(Alt_Real(Min)-N)
[0107] Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data;
[0108] The flight trajectory data is intercepted according to the timestamp. The data of each intercepted sequence includes six attributes: longitude difference, latitude difference, absolute altitude, speed along the longitude direction, speed along the latitude direction and vertical speed;
[0109] The intercepted sequence is made into a flight trajectory matrix based on the missing data, where the position of the missing data in the flight trajectory matrix is represented by 0, and the position of the non-missing data is represented by 1.
[0110] A generative adversarial network model module is used to construct a generative adversarial network model; wherein the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data;
[0111] Specifically, the generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module, and a fully connected module; the word embedding module includes a one-dimensional convolutional neural network layer; the downsampling module includes a one-dimensional convolutional neural network layer; the Transformer module includes a three-layer Transformer encoder; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer, and a skip connection layer; the fully connected module includes a three-layer fully connected neural network layer. The discriminator includes a one-dimensional convolutional network, a global pooling layer, and a fully connected layer.
[0112] The flight trajectory filling model module is used to train the adversarial network model based on the flight trajectory matrix set to obtain the flight trajectory filling model;
[0113] Specifically, in the flight trajectory filling model module, the binary cross entropy loss function is used to calculate the loss value of the discriminator output value, the mean square difference loss function is used to calculate the loss value of the generator output value, and the Adam optimizer is used to iteratively train the generator and discriminator.
[0114] The data output module is used to input the flight trajectory data to be filled into the flight trajectory filling model after preprocessing, so as to generate the filled flight trajectory data.
[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for filling missing points in flight trajectories based on generative adversarial networks, characterized in that: include: Step 1: Collect the flight trajectory data of the aircraft and preprocess it to obtain a flight trajectory matrix set; Step 2: construct a generative adversarial network model; wherein the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data; Step 3: training the adversarial network model based on the flight trajectory matrix set to obtain a flight trajectory filling model; Step 4: The flight trajectory data to be filled is pre-processed and then input into the flight trajectory filling model to generate filled flight trajectory data.
2. A method for filling missing points in a flight trajectory based on a generative adversarial network according to claim 1, characterized in that: The step 1 specifically includes: Step 1.1: Collect the flight trajectory data of the aircraft and extract the data of six attributes: longitude, latitude, altitude, horizontal speed, angle and vertical speed; Step 1.2: Based on the horizontal speed and the angle, the horizontal speed is decomposed into a speed Vx along the longitude direction and a speed Vy along the latitude direction according to the trigonometric function theorem, and the formula is as follows: Vx=V*cos(Head) Vy=V*sin(Head) Among them, V represents the horizontal speed, and Head represents the angle; Step 1.3: Using a difference calculation method, the longitude and the latitude are converted into a longitude difference and a latitude difference, and the formula is as follows: Lon_Diff=|Lon j+1 -Lon j |*100000 Lat_Diff=|Lat j+1 -Years j |*100000 Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j represents the latitude at time j; Step 1.4: Set the minimum altitude value to N, calculate the difference between the lowest altitude in the flight trajectory data and N, and subtract the difference from the altitude to get the relative altitude Alt: Alt=Alt_Real-(Alt_Real(Min)-N) Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data; Step 1.5: intercepting the flight trajectory data according to the timestamp, wherein the intercepted data of each sequence includes the longitude difference, the latitude difference, the absolute altitude, the speed in the longitude direction, the speed in the latitude direction and the vertical speed; Step 1.6: The intercepted sequence is made into a flight trajectory matrix according to the missing data, wherein the position of the missing data in the flight trajectory matrix is represented by 0, and the position without missing data is represented by 1.
3. The method for filling missing points in a flight trajectory based on a generative adversarial network according to claim 1, characterized in that: The generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module and a fully connected module; the word embedding module includes a one-dimensional convolutional neural network layer; the downsampling module includes a one-dimensional convolutional neural network layer; the Transformer module includes a three-layer Transformer encoder; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer and a skip connection layer; the fully connected module includes a three-layer fully connected neural network layer.
4. The method for filling missing points in a flight trajectory based on a generative adversarial network according to claim 1, characterized in that: The discriminator includes a three-layer one-dimensional convolutional network, a global pooling layer and a fully connected layer.
5. The method for filling missing points in a flight trajectory based on a generative adversarial network according to claim 1, characterized in that: In step 3, a binary cross entropy loss function is used to calculate the loss value of the discriminator output value, a mean square error loss function is used to calculate the loss value of the generator output value, and an Adam optimizer is used to iteratively train the generator and the discriminator.
6. A device for filling missing points in flight trajectories based on generative adversarial networks, characterized in that: include: A data acquisition preprocessing module is used to collect and preprocess the flight trajectory data of the aircraft to obtain a flight trajectory matrix set; A generative adversarial network model module, used to construct a generative adversarial network model; wherein the generative adversarial network model includes a generator and a discriminator; the generator is used to fill in the missing points of the flight trajectory matrix, and the discriminator is used to determine whether the flight trajectory data filled by the generator is the real flight trajectory data; A flight trajectory filling model module, used for training the adversarial network model based on the flight trajectory matrix set to obtain a flight trajectory filling model; The data output module is used to input the flight trajectory data to be filled into the flight trajectory filling model after preprocessing, so as to generate the filled flight trajectory data.
7. The device for filling missing points of a flight trajectory based on a generative adversarial network according to claim 6, characterized in that: The data acquisition preprocessing module is specifically used for: Collect the flight trajectory data of the aircraft and extract the data of six attributes: longitude, latitude, altitude, horizontal speed, angle and vertical speed; Based on the horizontal speed and the angle, the horizontal speed is decomposed into a speed Vx along the longitude direction and a speed Vy along the latitude direction according to the trigonometric function theorem, and the formula is as follows: Vx=V*cos(Head) Vy=V*sin(Head) Among them, V represents the horizontal speed, and Head represents the angle; The longitude and the latitude are converted into longitude difference and latitude difference by using the difference calculation method, and the formula is as follows: Lon_Diff=|Lon j+1 -Lon j |*100000 Lat_Diff=|Lat j+1 -Years j |*100000 Among them, Lon_Diff represents the longitude difference, Lat_Diff represents the latitude difference, Lon j+1 Indicates the longitude at time j+1, Lon j Indicates the longitude at time j, Lat j+1 Indicates the latitude at time j+1, Lat j represents the latitude at time j; Set the minimum altitude value to N, calculate the difference between the lowest altitude in the flight trajectory data and N, and subtract the difference from the altitude to get the relative altitude Alt: Alt=Alt_Real-(Alt_Real(Min)-N) Among them, Alt_Real represents the height of the flight trajectory data, and Alt_Real(Min) represents the minimum height of the flight trajectory data; The flight trajectory data is intercepted according to the timestamp, and each sequence of intercepted data includes the longitude difference, the latitude difference, the absolute altitude, the speed in the longitude direction, the speed in the latitude direction and the vertical speed; The intercepted sequence is made into a flight trajectory matrix according to the missing data, wherein the position of the missing data in the flight trajectory matrix is represented by 0, and the position without missing data is represented by 1.
8. The device for filling missing points of a flight trajectory based on a generative adversarial network according to claim 6, characterized in that: The generator includes a word embedding module, a downsampling module, a Transformer module, an upsampling module and a fully connected module; the word embedding module includes a one-dimensional convolutional neural network layer; the downsampling module includes a one-dimensional convolutional neural network layer; the Transformer module includes a three-layer Transformer encoder; the upsampling module includes a transposed one-dimensional convolutional neural network layer, a one-dimensional convolutional neural network layer and a skip connection layer; the fully connected module includes a three-layer fully connected neural network layer.
9. The device for filling missing points of a flight trajectory based on a generative adversarial network according to claim 6, characterized in that: The discriminator includes a one-dimensional convolutional network, a global pooling layer and a fully connected layer.
10. The device for filling missing points of a flight trajectory based on a generative adversarial network according to claim 6, characterized in that: In the flight trajectory filling model module, a binary cross entropy loss function is used to calculate the loss value of the discriminator output value, a mean square difference loss function is used to calculate the loss value of the generator output value, and an Adam optimizer is used to iteratively train the generator and the discriminator.