Track prediction method based on generative adversarial network and long-short term memory network

Through the GAN and LSTM-based track prediction method, the challenge of aircraft trajectory missing prediction under small sample conditions is solved. The small sample data is processed using polynomial fitting, TimeGAN and LSTM networks, and the accuracy of trajectory completion and prediction is achieved. Improved quality.

CN118982118AActive Publication Date: 2024-11-19BEIHANG UNIV

Patent Information

Application Number
CN202411071054.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-19
Estimated Expiration
2044-08-06

AI Technical Summary

Technical Problem

Under small sample conditions, aircraft trajectory missing prediction faces challenges. The existing methods require a lot of prior knowledge. Deep learning algorithms are difficult to adapt to complex and varied situations. Small sample learning problems are particularly prominent, especially the model structure and parameters caused by insufficient sample size. The inconsistency problem is difficult to solve.

Method used

Using a track prediction method based on generative adversarial network (GAN) and long and short-term memory network (LSTM), combining small sample data augmentation and intent identification through polynomial fitting interpolation, data normalization, TimeGAN network construction and LSTM network prediction. , process spatiotemporal correlation and historical information, and build GAN networks and LSTM networks to expand data and complete trajectories.

Benefits of technology

It improves the accuracy and robustness of trajectory prediction, solves the problems of missing trajectory data, long-term dependence, gradient disappearance and limited training data, and achieves fast and accurate trajectory completion, enhancing the generalization ability of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982118A_ABST
    Figure CN118982118A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-intention track prediction method based on a generative adversarial network and a long-short term memory network, belongs to the technical field of artificial intelligence time sequence prediction, and provides the multi-intention track prediction method based on the generative adversarial network and the long-short term memory network for a time sequence prediction problem. Aiming at the problems of track data missing, long-term dependence, gradient disappearance, limited training data and low real-time performance and generalization ability in a track prediction task, the invention provides a track prediction method for performing track complementation by utilizing space-time relevance and historical information between data through small sample data enhancement and small sample intention recognition. Meanwhile, the rapidity, robustness and accuracy of a trajectory prediction algorithm are ensured; a GAN network capable of expanding original data and an LSTM network capable of complementing tracks are constructed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of aviation and aerospace, and in particular to a track prediction method based on a generative adversarial network and a long short-term memory network. Background Art

[0002] The advancement of aviation technology has made aircraft trajectory completion algorithms a hot topic in the aviation field. This algorithm aims to solve the problem of missing trajectory data due to various reasons during flight, which is crucial to flight safety and data accuracy. At present, interpolation methods are widely used as basic technologies, including linear interpolation, polynomial interpolation, and spline interpolation, which infer and complete trajectory data by establishing mathematical models between known points. However, due to the complexity and variability of aircraft trajectories, these methods face challenges in dealing with nonlinear and complex systems.

[0003] In order to overcome the limitations of existing methods, researchers have proposed a variety of new algorithms. In response to the missing problem of ADS-B data, a multi-constraint condition discrimination model is established to screen missing points, and an improved neighboring algorithm and multiple interpolation method are used to repair the trajectory. In addition, based on the dynamic change of heading angle, researchers proposed an interpolation fitting completion algorithm for lost trajectory points. Sun Lishuang et al. proposed a neighborhood proportional completion algorithm, which uses the continuity characteristics of the neighborhood trajectory to determine the location of the completion point. Yu et al. studied a discrete Kalman filter trajectory prediction algorithm based on a neural acceleration estimator, which is suitable for military aircraft trajectory prediction and completion. In addition, deep learning methods, such as Bi-LSTM and Transformer, have also been applied to trajectory completion, using their nonlinear processing capabilities and self-attention mechanisms to improve prediction accuracy.

[0004] Despite the progress made in existing research, the field of trajectory completion still faces some challenges. Methods based on physical models require a lot of prior knowledge and experience, and are difficult to adapt to complex and changing actual situations. Although deep learning algorithms provide new solutions, the small sample learning problem remains a difficult problem because it is very difficult to obtain enough samples in practice. At present, the research on small sample learning focuses on three aspects: data enhancement, hypothesis space constraints, and search strategy adjustment. It aims to expand small sample data sets through prior knowledge and improve the generalization ability and credibility of the model. However, how to systematically design small sample learning strategies and how to deal with the inconsistency of model structure and parameters caused by too few samples are still difficult points that need to be further explored in the research. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and solve the problem of missing aircraft trajectory prediction under small sample conditions.

[0006] To achieve the above objectives, the present invention proposes a track prediction method based on a generative adversarial network and a long short-term memory network, comprising the following steps:

[0007] S1. Read the collected data and perform data cleaning.

[0008] S2. Perform polynomial fitting interpolation for missing trajectory data. Since the trajectory vector has a certain linear correlation with the time series, the known data can be used to construct a fitting polynomial to fill the missing trajectory data. The fitting polynomial is:

[0009]

[0010] Among them, Y m is the predicted dependent variable value, α0 is the intercept term, and α i is the regression coefficient, X i is the value of the ith independent variable in X, and N is the total number of data.

[0011] S3, normalize the three-dimensional trajectory feature data. The normalization formula for mapping the original data to the interval [0,1] is:

[0012]

[0013] Among them, the subscript x is the x-th dimension data in the trajectory vector, the subscript i is the i-th data, f′ is the value after mapping transformation, f is the value before transformation, and f′ xi Represents the i-th original data value f in the x-th dimension xi The value after normalization. F is the set of all vectors before transformation, that is, F = {f1, f2, ..., f i ,…,f N}, F x is the x-dimensional data in the set of all vectors before transformation, and N is the total number of data.

[0014] S4. Processing the small sample data set. The specific process is as follows: first, the small sample data set is segmented according to the intent, and the tracks with the same intent are extracted into a new data set while maintaining their time order, so that the segmented data set only contains data of a single type of intent.

[0015] S5. Build the TimeGAN network. The TimeGAN network consists of five parts: the generation network, the embedding network, the recovery network, the identification network, and the supervision network. The structure of each network is a 24-layer recurrent neural network plus a fully connected layer.

[0016] S6. Construct a loss function, and perform embedding training and supervised training in turn. By training these four networks, the data can maintain a distribution similar to the original after passing through the embedding network and the recovery network. First, the output representation of the input X and the noise Z after passing through different networks is given, where the noise Z is a random noise uniformly distributed from 0 to 1 with the same shape as the input data X. The output obtained by the input X after the embedding network is represented by H, and the output obtained by H after the recovery network is represented by Indicates that the output of H after the supervision network is Indicates that the output of H after the identification network is Y real Represents; the output of the noise Z after the generation network is expressed as express, The output obtained by the supervised network is express, The output obtained by restoring the network is express, The output obtained by the identification network is denoted by Y fake express, The output obtained by the identification network is denoted by Y efake express.

[0017] All the parameters of the five networks are stored in a list, and the parameter list of the embedded network is used v Indicates that the parameter list of the recovery network is represented by e r Indicates that the parameter list of the generated network is e g Indicates that the parameter list of the supervision network is expressed as e s Indicates that the parameter list of the identification network is represented by e d express.

[0018] The specific process is: Calculate the mean square loss during embedding training:

[0019]

[0020] Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The difference between the two is evaluated using the mean square error, and the adam optimizer is used to optimize e loss Update Parameters

[0021] Among them, θ1 is the parameter that needs to be updated, including e v and e r , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0022] Compute the mean squared loss during supervised training:

[0023]

[0024] Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The mean square error of the two is used as the loss of the supervision network, and the adam optimizer is used to optimize g loss Update Parameters

[0025] Among them, θ2 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0026] S7. Construct a joint loss function and perform joint training. First, calculate Y fake and Y fake The cross entropy loss for all-1 vectors of the same shape, Next, calculate Y efake and Y efake The cross entropy loss for all-1 vectors of the same shape, in Used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in (i.e., the output of the model), Y efake,i Yes efake The predicted value of the i-th sample in G. Gloss =g loss +g lossfr +g lossfre , using the adam optimizer to optimize G Gloss Update Parameters

[0027] Among them, θ3 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0028] Calculate Y real and Y real The cross entropy loss for all-1 vectors of the same shape, Next, calculate Y fake and Y fake The cross entropy loss of all zero vectors of the same shape, Next, calculate Y efake and Y efake The cross entropy loss of all zero vectors of the same shape, in Used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in (i.e., the output of the model), Y efake,i Yes efake The predicted value of the i-th sample in G. Dloss =g lossrr +g lossff +g lossffe , using the adam optimizer to optimize G Dloss Update Parameters

[0029] Among them, θ4 is the parameter that needs to be updated, including e d , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0030] In each training batch, θ3 is updated twice and θ4 is updated once.

[0031] S8. Construct a hidden layer LSTM network to predict the trajectory. The specific process is: construct the forget gate, update gate and output gate through the following formula:

[0032] Γ f =σ(W f [h t-1 ,x t ]+b f )

[0033] Γ u =σ(W u [h t-1 ,x t]+b u )

[0034] Γ a =σ(W a [h t-1 ,x t ]+b a )

[0035] Among them, Γ f , Γ u and Γ a They represent the forget gate, update gate and output gate respectively, σ is the Sigmoid function, W f , W u and W a is the weight coefficient matrix of the corresponding part, h is the output feature, x is the input feature, subscript t represents the current time, t-1 represents the previous time, b f 、b u and b a is the offset vector of the corresponding part. The candidate neurons are calculated by the following formula:

[0036]

[0037] Among them, W c is the weight coefficient matrix corresponding to this part, b c is the offset vector corresponding to this part. The updated neuron is calculated by the following formula:

[0038]

[0039] The output features at the current moment are calculated by the following formula:

[0040] h t =Γ a *tanh C t

[0041] S9, construct the output fully connected layer to obtain the three-dimensional coordinate prediction trajectory. The specific process is: construct the fully connected layer through the following formula:

[0042] y f =W f ·h t +b f

[0043] Among them, W f is the weight matrix of the fully connected layer, b f is the bias of the fully connected layer, y f It is the output scalar of the fully connected layer, which contains the predicted trajectory information of one time step. During prediction, the trajectory points of one time step in one dimension are predicted each time, and the final output is the predicted trajectory of the three-dimensional coordinates corresponding to one time step.

[0044] Compared with the prior art, the advantages and beneficial effects of the present invention are: in order to address the problems of missing trajectory data, long-term dependence, gradient disappearance, limited training data, poor real-time and generalization capabilities in trajectory prediction tasks, a trajectory prediction method of small sample data enhancement and small sample intention recognition, and trajectory completion using the spatiotemporal correlation and historical information between data is proposed, while ensuring the speed, robustness and accuracy of the trajectory prediction algorithm; a GAN network that can expand the original data and an LSTM network that completes the trajectory are constructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention is a flow chart of a track prediction method based on a generative adversarial network and a long short-term memory network.

[0046] Figure 2 This is the LSTM network operation structure diagram of the present invention.

[0047] Figure 3 This is a structural diagram of the LSTM network model of the present invention.

[0048] Figure 4 This is a graph showing the prediction results of the original data of an example of the present invention.

[0049] Figure 5 This is a diagram of the prediction results of the expanded data of the example of the present invention. DETAILED DESCRIPTION

[0050] To further illustrate the features of the present invention, please refer to the following detailed description of the present invention and the attached Figure 1-5 The accompanying drawings are for reference and illustration purposes only and are not intended to limit the scope of protection of the present invention.

[0051] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings:

[0052] like Figure 1 As shown, this embodiment discloses a track prediction method based on a generative adversarial network and a long short-term memory network of the present invention, comprising the following steps:

[0053] S1, read the collected data and perform data cleaning. Data collection part: FlightADSB is a platform that provides real-time tracking services for global flight trajectories. Users can obtain real-time tracking maps of aircraft, aircraft flight trajectory playback, flight trajectory data download, flight status, ADS-B equipment application and aircraft picture display and other information services through FlightADSB. Therefore, the example part of the present invention selects three tracks in FlightADSB corresponding to three intentions, and the flight number is HO1702 (Guiyang Longdongbao-Nanjing Lukou) with an intention label mapping of 1, CZ6796 (Guiyang Longdongbao-Sanya Phoenix) with an intention label mapping of 2, and ZH8210 (Guiyang Longdongbao-Shenzhen Baoan) with an intention label mapping of 3. Ten complete trajectories are taken and 10 incomplete trajectories are intercepted as training data sets to reflect the effect of this method in solving the small sample problem under the condition of insufficient data and missing data. In addition, in order to reflect the actual conditions and the limitations of current sensor technology, the data that can be obtained in real time in reality is retained, including flight timestamps, geographic coordinates, altitude and other information. Data cleaning part: including deletion of duplicate values, consistency processing, and outlier detection and processing operations.

[0054] S2. Perform polynomial fitting interpolation for missing trajectory data. In real scenarios, data may be lost during the collection, transmission and analysis process, and the integrity of the data set is crucial to accurately identify the true intention of the flying target. Therefore, it is very necessary to effectively fill in the missing parts of the trajectory. Since the trajectory vector has a certain linear correlation with the time series, the known data can be used to construct a fitting polynomial to fill in the missing trajectory data. The fitting polynomial is:

[0055]

[0056] Among them, Y m The data set contains one or more missing attribute values. mis ,

[0057] X i (i=1,…,N) is the instance X in the data set without any missing attribute values obs .

[0058] S3. Normalize the three-dimensional trajectory feature data. Data normalization can eliminate the impact of data dimension and improve network convergence efficiency. The normalization formula for mapping the original data to the interval [0,1] is:

[0059]

[0060] Among them, the subscript x is the x-th dimension data in the trajectory vector, the subscript i is the i-th data, f′ is the value after mapping transformation, f is the value before transformation, and F is the set of all vectors before transformation, that is, F={f1,f2,…,f i ,…,f N}, N is the total number of data.

[0061] S4. Processing the small sample data set. The specific process is as follows: first, the small sample data set is segmented according to the intent, and the tracks with the same intent are extracted into a new data set while maintaining their time order, so that the segmented data set only contains data of a single type of intent.

[0062] S5. Build the TimeGAN network. The TimeGAN network consists of five parts: the generation network, the embedding network, the recovery network, the identification network, and the supervision network. The structure of each network is a 24-layer recurrent neural network plus a fully connected layer.

[0063] S6. Construct a loss function, and perform embedding training and supervised training in turn. By training these four networks, the data can maintain a distribution similar to the original after passing through the embedding network and the recovery network. First, the output representation of the input X and the noise Z after passing through different networks is given, where the noise Z is a random noise uniformly distributed from 0 to 1 with the same shape as the input data X. The output obtained by the input X after the embedding network is represented by H, and the output obtained by H after the recovery network is represented by Indicates that the output of H after the supervision network is Indicates that the output of H after the identification network is Y real Represents; the output of the noise Z after the generation network is expressed as express, The output obtained by the supervised network is express, The output obtained by restoring the network is express, The output obtained by the identification network is denoted by Y fake express, The output obtained by the identification network is denoted by Y efake express.

[0064] All the parameters of the five networks are stored in a list, and the parameter list of the embedded network is used v Indicates that the parameter list of the recovery network is represented by e r Indicates that the parameter list of the generated network is e g Indicates that the parameter list of the supervision network is expressed as e s Indicates that the parameter list of the identification network is represented by e d express.

[0065] The specific process is: Calculate the mean square loss during embedding training:

[0066]

[0067] Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The difference between the two is evaluated using the mean square error, and the adam optimizer is used to optimize e loss Update Parameters

[0068] Among them, θ1 is the parameter that needs to be updated, including e v and e r , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0069] Compute the mean squared loss during supervised training:

[0070]

[0071] Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The mean square error of the two is used as the loss of the supervision network, and the adam optimizer is used to optimize g loss Update Parameters

[0072] Among them, θ2 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0073] S7, construct joint loss function, network joint training. First calculate Y fake and Y fake The cross entropy loss for all-1 vectors of the same shape, Next, calculate Y efake and Y efakeThe cross entropy loss for all-1 vectors of the same shape, in Used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in (i.e., the output of the model), Y efake,i Yes efake The predicted value of the i-th sample in G. Gloss =g loss +g lossfr +g lossfre , using the adam optimizer to optimize G Gloss Update Parameters

[0074] Among them, θ3 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0075] Calculate Y real and Y real The cross entropy loss for all-1 vectors of the same shape,

[0076] Next, calculate Y fake and Y fake The cross entropy loss of all zero vectors of the same shape, Next, calculate Y efake and Y efake The cross entropy loss of all zero vectors of the same shape, in Used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in (i.e., the output of the model), Y efake,i Yes efake The predicted value of the i-th sample in G. Dloss =g lossrr +g lossff +g lossffe , using the adam optimizer to optimize G Dloss Update Parameters

[0077] Among them, θ4 is the parameter that needs to be updated, including e d, α represents the learning rate, which is a hyperparameter used to control the step size of each update. It is an unbiased estimate of the gradient and is used to guide the direction of parameter update. is the variance of the unbiased estimate of the gradient. ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

[0078] In each training batch, θ3 is updated twice and θ4 is updated once.

[0079] S8. Construct a hidden layer LSTM network to predict the trajectory. The specific process is: construct the forget gate, update gate and output gate through the following formula:

[0080] Γ f =σ(W f [h t-1 ,x t ]+b f )

[0081] Γ u =σ(W u [h t-1 ,x t ]+b u )

[0082] Γ a =σ(W a [h t-1 ,x t ]+b a )

[0083] Among them, Γ f , Γ u and Γ a They represent the forget gate, update gate and output gate respectively, σ is the Sigmoid function, W f , W u and W a is the weight coefficient matrix of the corresponding part, h is the output feature, x is the input feature, subscript t represents the current time, t-1 represents the previous time, b f 、b u and b a is the offset vector of the corresponding part. The candidate neurons are calculated by the following formula:

[0084]

[0085] Among them, W c is the weight coefficient matrix corresponding to this part, b c is the offset vector corresponding to this part. The updated neuron is calculated by the following formula:

[0086]

[0087] The output features at the current moment are calculated by the following formula:

[0088] h t =Γ a *tanh C t

[0089] The operation structure diagram of the above LSTM network is as follows Figure 2 As shown in the figure. The LSTM model is divided into three parts: historical trajectory input layer, hidden layer and output layer. The network input node input = (100,3), output node output = 3, where 100 represents the historical trajectory step length; 3 represents the feature dimension. The LSTM network model structure diagram is shown in the figure. Figure 3 shown.

[0090] S9, construct the output fully connected layer to obtain the three-dimensional coordinate prediction trajectory. The specific process is: construct the fully connected layer through the following formula:

[0091] y f =W f ·h t +b f

[0092] Among them, W f is the weight matrix of the fully connected layer, b f is the bias of the fully connected layer, y f It is the output scalar of the fully connected layer, which contains the predicted trajectory information of one time step. During prediction, the trajectory points of one time step in one dimension are predicted each time, and the final output is the predicted trajectory of the three-dimensional coordinates corresponding to one time step.

[0093] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0094] The following is a simulation example. The present invention will use an example to illustrate the effect of the proposed method. This case shows the advantages of the proposed method.

[0095] First, the three tracks in FlightADSB are mapped to three intent labels. HO17002 is mapped to intent Figure 1 , CZ6796 is mapped to Figure 2 , ZH8210 is mapped to Figure 3 . Secondly, we build TimeGAN to expand the generated data.

[0096] The generated results in the example are analyzed using the PCA package in Scikit-Learn. The pca.components_ attribute of the PCA model gives the eigenvector corresponding to each principal component. In order to reduce the dimension, usually only a part of the principal components is retained, which means that all variances will not be considered. The proportions of PCA principal components for various purposes and their corresponding original features are shown in the following table. The numbers outside the brackets in the table represent the number of features of the data corresponding to the principal component, and the numbers in the brackets represent the contribution size and direction of each original feature in the principal component.

[0097] Table 1. The proportion and characteristics of the principal components of TGAN with different intentions

[0098]

[0099] As can be seen in Table 1, the principal components of the generated data show significant similarities in characteristics with the principal components of the original data. The principal components of the generated data are similar to those of the original data in terms of direction and proportion, which indicates that the generated data retains similar characteristics to the original data in the direction of the principal components. This property indicates to the generation algorithm that the generated data is consistent with the distribution of the original data in key aspects.

[0100] Combine the augmented data with the original data and add intent labels according to different intents. Send it to the LSTM network for trajectory prediction model training. The original data set predicts the trajectory as follows Figure 4 As shown in Figure 2, the RMSE value of the original trajectory and the predicted trajectory is 79.687. The predicted trajectory of the expanded dataset is shown in Figure 2. Figure 5 As shown, the RMSE value of the original trajectory and the predicted trajectory is 64.189. Compared with the original data set, the method adopted by the present invention improves the accuracy of trajectory prediction on three-dimensional data.

Claims

1. A track prediction method based on generative adversarial network and long short-term memory network, characterized in that: The steps include: S1, read the collected data and perform data cleaning; S2. Perform polynomial fitting interpolation for missing trajectory data; use known data to construct a fitting polynomial to fill the missing trajectory data; the fitting polynomial is: Among them, Y m is the predicted dependent variable value, α0 is the intercept term, and α i is the regression coefficient, X i is the value of the ith independent variable in X, and N is the total number of data; S3, normalize the three-dimensional trajectory feature data; the normalization formula for mapping the original data to the interval [0,1] is: Among them, the subscript x is the x-th dimension data in the trajectory vector, the subscript i is the i-th data, f′ is the value after mapping transformation, f is the value before transformation, and f′ xi Represents the i-th original data value f in the x-th dimension xi The value after normalization; F is the set of all vectors before transformation, that is, F = {f1, f2, ..., f i ,…,f N }, F x is the x-th dimension data in the set of all vectors before transformation, and N is the total number of data; S4, processing the small sample data set; segmenting the small sample data set according to the intention, extracting the tracks with the same intention into a new data set while keeping their time order, so that the segmented data set only contains data of a single type of intention; S5. Build the TimeGAN network. The TimeGAN network consists of five parts: generation network, embedding network, recovery network, identification network, and supervision network. The structure of each network is a 24-layer recurrent neural network plus a fully connected layer. S6. Construct a loss function and perform embedding training and supervised training in sequence. S7, construct a joint loss function and perform joint training; S8, construct a hidden layer LSTM network to predict the trajectory; S9. Construct an output fully connected layer to obtain the three-dimensional coordinate prediction trajectory.

2. The track prediction method based on generative adversarial network and long short-term memory network according to claim 1, characterized in that: In step S6, the output representations of the input X and noise Z after passing through different networks are first given; wherein the noise Z is a random noise uniformly distributed from 0 to 1 with the same shape as the input data X; the output obtained by the input X after the embedding network is represented by H, and the output obtained by H after the restoration network is represented by Indicates that the output of H after the supervision network is Indicates that the output of H after the identification network is Y real Represents; the output of the noise Z after the generation network is expressed as express, The output obtained by the supervised network is express, The output obtained by restoring the network is express, The output obtained by the identification network is denoted by Y fake express, The output obtained by the identification network is denoted by Y efake express.

3. The track prediction method based on generative adversarial network and long short-term memory network according to claim 2, characterized in that: All parameters of the five networks are stored in a list, and the parameter list of the embedded network is used v Indicates that the parameter list of the recovery network is represented by e r Indicates that the parameter list of the generated network is e g Indicates that the parameter list of the supervision network is expressed as e s Indicates that the parameter list of the identification network is represented by e d express.

4. A track prediction method based on generative adversarial network and long short-term memory network according to claim 2 or 3, characterized in that: Calculate the mean squared loss during embedding training: Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The i-th component in ; the mean square error is used to evaluate the difference between the two, and the adam optimizer is used to optimize e loss Update Parameters Among them, θ1 is the parameter that needs to be updated, including e v and e r , α represents the learning rate, which is a hyperparameter used to control the step size of each update; It is an unbiased estimate of the gradient, which is used to guide the direction of parameter update; is the variance of the unbiased estimate of the gradient; ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

5. The method for track prediction based on generative adversarial network and long short-term memory network according to claim 4, characterized in that: Compute the mean squared loss during supervised training: Where N is the total number of data, subscript i is the i-th component in the data, and X i is the i-th component in the input data, for The i-th component in g is used; the mean square error of the two is used as the loss of the supervision network, and the adam optimizer is used to optimize g loss Update Parameters Among them, θ2 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update; It is an unbiased estimate of the gradient, which is used to guide the direction of parameter update; is the variance of the unbiased estimate of the gradient; ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

6. The track prediction method based on generative adversarial network and long short-term memory network according to claim 1, characterized in that: In step S7, first calculate Y fake and Y fake The cross entropy loss for all-1 vectors of the same shape, Next, calculate Y efake and Y efake The cross entropy loss for all-1 vectors of the same shape, in It is used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in Y efake,i Yes efake The predicted value of the i-th sample in G Gloss =g loss +g lossfr +g lossfre , using the adam optimizer to optimize G Gloss Update Parameters Among them, θ3 is the parameter that needs to be updated, including e g and e s , α represents the learning rate, which is a hyperparameter used to control the step size of each update; It is an unbiased estimate of the gradient, which is used to guide the direction of parameter update; is the variance of the unbiased estimate of the gradient; ∈ is a small positive number used for numerical stability to prevent the denominator from being zero.

7. The method for track prediction based on generative adversarial network and long short-term memory network according to claim 6, characterized in that: Calculate Y real and Y real The cross entropy loss for all-1 vectors of the same shape, Next, calculate Y fake and Y fake The cross entropy loss of all zero vectors of the same shape, Next, calculate Y efake and Y efake The cross entropy loss of all zero vectors of the same shape, in It is used to map any real number to the interval (0,1), N is the total number of samples, Y fake,i Yes fake The predicted value of the i-th sample in (i.e., the output of the model), Y efake,i Yes efake The predicted value of the i-th sample in G Dloss =g lossrr +g lossff +g lossffe , using the adam optimizer to optimize G Dloss Update Parameters Among them, θ4 is the parameter that needs to be updated, including e d , α represents the learning rate, which is a hyperparameter used to control the step size of each update; It is an unbiased estimate of the gradient, which is used to guide the direction of parameter update; is the variance of the unbiased estimate of the gradient; ∈ is a very small positive number used for numerical stability to prevent the denominator from being zero; in each training batch, θ3 is updated twice and θ4 is updated once.

8. The method for track prediction based on generative adversarial network and long short-term memory network according to claim 1, characterized in that: In step S8, the forget gate, update gate and output gate are constructed by the following formula: C f =σ(W f [h t-1 ,x t ]+b f ) C u =σ(W u [h t-1 ,x t ]+b u ) C a =σ(W a [h t-1 ,x t ]+b a ) Among them, Γ f , Γ u and Γ a They represent the forget gate, update gate and output gate respectively, σ is the Sigmoid function, W f , W u and W a is the weight coefficient matrix of the corresponding part, h is the output feature, x is the input feature, subscript t represents the current time, t-1 represents the previous time, b f 、b u and b a is the offset vector of the corresponding part.

9. The method for track prediction based on generative adversarial network and long short-term memory network according to claim 8, characterized in that: The candidate neurons are calculated by the following formula: Among them, W c is the weight coefficient matrix corresponding to this part, b c is the offset vector corresponding to this part; the updated neuron is calculated by the following formula: The output features at the current moment are calculated by the following formula: h t =Γ a *tanh C t 。 10. The method for track prediction based on generative adversarial network and long short-term memory network according to claim 1, characterized in that: In step S9, a fully connected layer is constructed by the following formula: y f =W f ·h t +b f Among them, W f is the weight matrix of the fully connected layer, b f is the bias of the fully connected layer, y f It is the output scalar of the fully connected layer, which contains the predicted trajectory information of one time step. When predicting, it predicts the trajectory points of one time step in one dimension each time, and finally outputs the predicted trajectory of the three-dimensional coordinates corresponding to one time step.

Citation Information

Patent Citations

  • Aircraft trajectory prediction method based on long short-term memory network

    CN114048889A

  • Aircraft trajectory prediction system based on semi-supervised learning and neural network

    CN117332327A

Cited By

  • Intelligent track prediction method based on Transform-LSTM

    CN119785630A