Flight trajectory prediction method based on multi-channel-frequency

Abstract

The invention provides a multi-channel-frequency-based flight path prediction method, and relates to the technical field of flight path prediction. The method comprises the following steps: acquiring track sequence data, extracting time patterns of a plurality of features of the track sequence data, fusing and embedding, and acquiring track sequence feature embedding; embedding and extracting low-frequency and high-frequency information based on track sequence features; respectively inputting the low-frequency and high-frequency information into a frequency feature extraction module, and extracting low-frequency and high-frequency key information; converting the low-frequency and high-frequency key information into low-frequency and high-frequency signals; based on the low-frequency and high-frequency signals, acquiring a time-domain trajectory sequence, and intercepting a final point as a predicted trajectory point; and performing iterative calculation to obtain a plurality of prediction trajectory points, and sequentially connecting the prediction trajectory points to complete flight trajectory prediction. According to the method, the perception capability of flight state change can be effectively enhanced, the accuracy of flight trajectory prediction is improved, and the space-time coherence of the trajectory is guaranteed; in general, flight trajectory prediction under more scene types and more complex flight states can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of flight trajectory prediction, and in particular to a flight trajectory prediction method based on multi-channel-frequency. Background Art

[0002] In air traffic management, flight trajectory prediction is an important part. By providing accurate flight trajectory information to air traffic controllers, airspace conditions can be better monitored, potential conflicts can be discovered in a timely manner, and the possibility of flight delays can be predicted in advance. In actual aviation operations, accurate and effective flight trajectory prediction is essential to ensure flight safety and optimize control operations. The multi-channel-frequency flight trajectory prediction method can improve the accuracy of flight trajectory prediction and provide a basis for airlines to formulate more reasonable flight plans.

[0003] Existing flight trajectory prediction methods can be divided into three types, namely dynamic modeling methods, state estimation methods and machine learning methods. The dynamic modeling method establishes kinematic equations to simulate flight patterns and predict future motion trajectories. The state estimation method propagates and estimates subsequent motion states based on aircraft properties (such as position, velocity, flight speed, etc.). Traditional machine learning methods use a large amount of historical trajectory data to learn the basic patterns and characteristics of aircraft motion, but require experts to manually design features. With the development of deep learning, deep neural networks have gradually emerged, which can automatically learn complex hierarchical features from input data to represent input data without the need for manual feature engineering. The deep learning prediction methods commonly used at this stage have the following problems: First, the model ignores the flight trajectory trend and it is difficult to capture changes in flight status; second, it ignores the time pattern of each variable and it is difficult to capture the correlation between flight variables. Summary of the invention

[0004] In order to solve the above technical problems existing in the prior art, the present invention aims to propose a flight trajectory prediction method that captures the overall trend of flight and the changes in flight motion details, and integrates the time dependency in flight sequence data.

[0005] Specifically, the present invention provides a flight trajectory prediction method based on multi-channel-frequency, and the technical solution specifically includes the following steps:

[0006] Step S1: Obtain trajectory sequence data of a single flight including multiple features, wherein each feature corresponds to a channel;

[0007] Step S2: extract the time pattern of each feature of the trajectory sequence data through the MLP (Multilayer Perceptron) network and fuse them to obtain the trajectory sequence feature representation, and then perform embedded expression to obtain the trajectory sequence feature embedding;

[0008] Step S3: embedding the trajectory sequence features as input, extracting low-frequency information through the first convolutional neural network, and extracting high-frequency information through the second convolutional neural network;

[0009] Step S4: inputting the low-frequency information and the high-frequency information into a frequency feature extraction module respectively to extract low-frequency key information and high-frequency key information; wherein the frequency feature extraction module includes three independent Mamba blocks;

[0010] Step S5: converting the low-frequency key information into a low-frequency signal, and converting the high-frequency key information into a high-frequency signal;

[0011] Step S6: using the low-frequency signal and the high-frequency signal as input, obtaining the time-domain trajectory sequence of the trajectory sequence data through inverse wavelet transform, and intercepting the last point of the time-domain trajectory sequence as the predicted trajectory point;

[0012] Step S7: Based on the trajectory sequence data and the predicted trajectory points, iteratively calculate new predicted trajectory points; connect all the predicted trajectory points in sequence as the result of flight trajectory prediction.

[0013] Furthermore, before step S1, the following steps are also included:

[0014] A trajectory data set is obtained, and missing and duplicate trajectory data are deleted; wherein the trajectory data set includes trajectory sequence data of multiple flights.

[0015] Furthermore, after step S1, the following steps are further included:

[0016] Unify the time granularity of trajectory sequence data through piecewise linear interpolation and resampling;

[0017] The trajectory sequence data of different lengths are divided into slices of the same size X through the sliding window operation 1:t ; where t is the slice length.

[0018] Preferably, step S2 specifically includes:

[0019] Each slice X 1:t After transposition, the full connection layer and the ReLU activation layer are input in sequence to obtain the corresponding time mode D.

[0020] D=ReLu(Linear(X 1:t ) T );

[0021] In the formula, Linear(·) represents the fully connected operation;

[0022] Transpose the time pattern D and slice it with X 1:t Add together to obtain the trajectory sequence feature representation D',

[0023] D′=DT +X 1:t ;

[0024] The trajectory sequence feature representation D′ is embedded by the Embedding method.

[0025] E = Embedding(D′);

[0026] Where E is the trajectory sequence feature embedding.

[0027] Preferably, step S3 specifically includes:

[0028] E low =Conv L (E);

[0029] E high =Conv H (E);

[0030] In the formula, Conv L (·) and Conv H (·) represent the first convolutional neural network and the second convolutional neural network respectively, E low and E high They are low-frequency information and high-frequency information respectively.

[0031] Preferably, step S4 specifically includes:

[0032] The low frequency information E low Input three independent Mamba blocks respectively, and extract low-frequency key information A through the attention mechanism low ,

[0033] Q low =Mamba0(E low );

[0034] K low =Mamba1(E low );

[0035] V low =Mamba2(E low );

[0036] A low =Q low ⊙K low ⊙V low ;

[0037] In the formula, Q low , K low and V low All of them are intermediate calculations; Mamba0(·), Mamba1(·) and Mamba2(·) represent three independent Mamba blocks respectively;

[0038] The high frequency information E high Input three independent Mamba blocks respectively, and extract high-frequency key information A through the attention mechanism high ,

[0039] Q high =Mamba3(E high );

[0040] K high =Mamba4(E high );

[0041] V high =Mamba5(E high );

[0042] A high =Q high ⊙K high ⊙V high ;

[0043] In the formula, Q high , K high and V high All of them are intermediate calculations; Mamba3(·), Mamba4(·) and Mamba5(·) represent three independent Mamba blocks respectively.

[0044] Preferably, step S4 further includes:

[0045] Mamba0(·) and Mamba3(·) are the same Mamba block.

[0046] Preferably, step S5 specifically includes:

[0047] For low frequency key information A low Convolution operation and linear processing are performed in sequence to obtain the low-frequency signal F low ,

[0048] F low =Linear(Conv l (A low ));

[0049] In the formula, Conv l (·) indicates the third convolutional layer;

[0050] For high frequency key information A high Convolution operation and linear processing are performed in sequence to obtain the high-frequency signal F high ,

[0051] F high =Linear(Conv h (Ahigh ));

[0052] In the formula, Conv h (·) indicates the fourth convolutional layer.

[0053] Preferably, step S6 specifically includes:

[0054] Y 1:t+1 =IDWT(F low ,F high )

[0055] Where Y 1:t+1 represents the time domain trajectory sequence, IDWT(·) represents the inverse wavelet transform;

[0056] Intercept the last point Y t+1 is the trajectory prediction point P.

[0057] The technical solution provided by the present invention can effectively enhance the ability to perceive changes in flight status and improve the accuracy of flight trajectory prediction; and by converting multi-frequency features into time domain trajectory sequences through inverse wavelet transform, and combining iterative prediction to generate a continuous trajectory point sequence, the spatiotemporal coherence of the trajectory can be guaranteed; in short, the technical solution provided by the present invention can realize flight trajectory prediction in more scene types and more complex flight conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the overall model structure of the multi-channel-frequency based flight trajectory prediction in the present invention.

[0059] Figure 2 This is a detailed architecture diagram of the Mamba module in the present invention.

[0060] Figure 3 For the ascending stage under the real data set, the present invention provides a comparison chart of the effect of flight trajectory prediction with the existing prediction method.

[0061] Figure 4 For the descent stage under a real data set, the present invention provides a comparison chart of the effect of flight trajectory prediction with the existing prediction method.

[0062] Figure 5 For the cruise phase under a real data set, the present invention provides a comparison chart of the effect of flight trajectory prediction with the existing prediction method.

[0063] Figure 6 For the turning stage under a real data set, the present invention provides a comparison chart of the effect of flight trajectory prediction with the existing prediction method. DETAILED DESCRIPTION

[0064] The technical solution provided by the present invention will be further described in detail below with reference to the accompanying drawings.

[0065] like Figure 1 As shown, the flight trajectory prediction is performed based on the collected trajectory data set, including the following steps:

[0066] Step S1: preprocessing the trajectory sequence data;

[0067] Step S2: using an MLP network to extract and fuse the temporal patterns of each channel;

[0068] Step S3: Two convolutional neural networks are used to preliminarily extract low-frequency information E low and high frequency information E high ;

[0069] Step S4: Use the Mamba-based frequency feature extraction module to further extract low-frequency key information A low and high frequency key information A high ;

[0070] Step S5: Use convolutional neural network to convert high-frequency and low-frequency key information into low-frequency signal F low and high frequency signal F high ;

[0071] Step S6: Use inverse wavelet transform to transform the frequency signal F low and F high Convert it into a time domain trajectory sequence and take the last point as the predicted trajectory point.

[0072] Step S1 specifically includes:

[0073] Step S11: Obtain the trajectory data set, remove missing and duplicate data, and extract trajectory sequence data individually according to the flight number to ensure data integrity; wherein the trajectory sequence data includes multiple features such as longitude, latitude, altitude and speed, and each feature corresponds to a channel;

[0074] Step S12: performing piecewise linear interpolation and resampling operations on the trajectory data at intervals of 10 seconds to unify the time granularity;

[0075] Step S13: Slice the trajectory through a sliding window operation, and divide the trajectory sequences of different lengths into slices X of the same size. 1:t , where t is the slice length.

[0076] Step S2 specifically includes:

[0077] Step S21: Input the trajectory into slice X 1:tPerform a transposition operation, pass through the fully connected layer and the ReLU activation layer in sequence, and extract the temporal pattern D of different features in the input sequence:

[0078] D=ReLu(Linear(X 1:t ) T );

[0079] Step S22: Transpose the trajectory sequence slice time pattern D and combine it with the original input trajectory sequence slice X 1:t Add, retain and fuse the original channel time pattern to obtain the trajectory sequence feature representation D':

[0080] D'=D T +X 1:t ;

[0081] Step S3 specifically includes:

[0082] Step S31: Slice the trajectory sequence D′ after fusion of channel time patterns and embed it into 64 dimensions using the Embedding method to obtain the semantic representation E:

[0083] E = Embedding(D′);

[0084] Step S32: The embedding E is passed through two convolutional neural networks with different receptive field sizes, a convolution kernel size of 1×1 and a step size of 1. L Focus on local areas and extract the basic pattern features E that evolve slowly in the trajectory low , convolution kernel size 3×3, stride 1, padding 1 convolution Conv H It is used to capture relatively long-distance dependencies in trajectory sequences and extract fast-changing detail features. high :

[0085] E low =Conv L (E);

[0086] E high =Conv H (E);

[0087] Step S4 specifically includes:

[0088] Step S41: The low frequency information E low and high frequency information E high After passing through three Mamba blocks respectively. Combined with the attention mechanism, three different Mamba blocks are used as query (Q), key (K) and value (V) components to extract their respective key information. Finally, the respective Q, K, V Hadamard products are applied on the low-frequency and high-frequency branches to extract finer-grained local features A lowand richer contextual information A high :

[0089] Q = Mamba0(L);

[0090] K = Mamba1(L);

[0091] V = Mamba2(L);

[0092] A=Q⊙K⊙V;

[0093] Where L = E low or E high , A=A low or A high , Q, K, V are only used to generate the attention matrix to extract the contextual key information A, ⊙ represents the Hadamard multiplication.

[0094] The Mamba block has a detailed architecture as follows Figure 2 As shown. Among them, X t Indicates input, Y t represents the output, A1 represents the structured state matrix, B t , C t Both represent learnable parameters, and Δt represents the step size used to discretize the continuous sequence.

[0095] Step S42: The Mamba0 module computing unit that shares the high-frequency information and the low-frequency information is used to enhance the interaction between the high-frequency information and the low-frequency information:

[0096] Mamba0 = SharedMamba;

[0097] Step S5: Extract the low-frequency key information A low and high frequency key information A low After passing through independent convolutional layers and linear layers respectively, the low-frequency signal F is obtained. low and high frequency signal F high :

[0098] F low =Linear(Conv l (A low ));

[0099] F high =Linear(Conv h (A high ));

[0100] Among them, Conv l and Conv h The convolution kernel size is 3×3, the stride is 2, and the padding is 1;

[0101] Step S6 specifically includes:

[0102] Step S61: Use inverse wavelet transform to transform the frequency signal F low and F high Convert to time domain trajectory sequence Y 1:t+1 :

[0103] Y 1:t+1 =IDWT(F low ,F high );

[0104] Step S62: intercept the last point as the trajectory prediction point P:

[0105] P=Y t+1 ;

[0106] Step S63: Take the predicted point P as the trajectory node at the next moment, and form a complete predicted trajectory sequence T through multiple iterations of prediction. L :

[0107] T L = {Y t+1 ,Y t+2 ,…,Y t+n}

[0108] Where n is the length of the predicted trajectory sequence.

[0109] like Figure 3 As shown, the X, Y, and Z axes in the three-dimensional coordinates represent longitude, latitude, and altitude, respectively. The legends on the upper left are: input sequence Inputs, true prediction sequence GrouundTruth, WTFTP model prediction sequence, model prediction sequence of the present invention, LSTM (Long Short-Term Memory) model prediction sequence, and CNNLSTM (Convolutional Neural Networks Long Short-Term Memory) model prediction sequence. Real-world flight data sets are used for model training and verification, and the prediction performance of each model in different spatial locations is used for display and evaluation. Figure 3 , 4 As shown in Figures 5 and 6, by comparing the experimental results, it can be seen that the prediction method of the present invention shows higher accuracy and continuity under different flight conditions, whether in the flight's ascending phase, descending phase, cruising phase or turning phase, and has significant advantages over traditional methods.

[0110] The present invention not only solves the core problems in flight trajectory prediction, but also significantly improves the accuracy and consistency of the model, and is suitable for trajectory prediction under complex flight conditions. Aiming at the key problems of existing flight trajectory prediction technology, a prediction method based on multi-frequency and multi-channel fusion is proposed, and significant technical effects are achieved through targeted technical means. First, traditional models often ignore the overall trend of flight trajectories and are difficult to capture changes in flight status. The present invention decomposes and models low-frequency and high-frequency features in trajectory data through a multi-frequency Mamba module. Low-frequency features are used to capture the overall trend of flight trajectories, and high-frequency features are used to identify rapid changes in flight details, thereby effectively enhancing the model's perception of flight state changes and improving the accuracy of trajectory prediction. In addition, existing models often ignore the time pattern of each variable, making it difficult to capture the correlation between flight variables. The present invention extracts and fuses the time pattern of each channel through an MLP network to fully identify the time dependency between variables. Finally, traditional prediction methods have deficiencies in the spatiotemporal continuity of trajectory prediction, which affects the overall prediction effect. The present invention converts multi-frequency features into a time-domain trajectory sequence through an inverse wavelet transform, and generates a continuous trajectory point sequence in combination with iterative prediction to ensure the spatiotemporal consistency of the trajectory.

[0111] In summary, it can be seen that the technical solution provided by the present invention can effectively enhance the ability to perceive changes in flight status and improve the accuracy of flight trajectory prediction; and by converting multi-frequency features into time domain trajectory sequences through inverse wavelet transform, and combining iterative prediction to generate a continuous trajectory point sequence, the spatiotemporal coherence of the trajectory can be guaranteed; in short, the technical solution provided by the present invention can realize flight trajectory prediction in more scene types and more complex flight conditions.

[0112] Furthermore, on the basis of achieving the above-mentioned beneficial effects, each preferred scheme also achieves the following beneficial effects: deleting missing and duplicate trajectory data helps to improve the accuracy and precision of flight trajectory prediction; unifying the time granularity and slicing trajectory sequence data is conducive to improving the accuracy and efficiency of flight trajectory prediction.

Claims

1. A flight trajectory prediction method based on multi-channel-frequency, characterized in that: The following steps are involved: Step S1: Obtain trajectory sequence data of a single flight including multiple features, wherein each feature corresponds to a channel; Step S2: extract the time pattern of each feature of the trajectory sequence data through the MLP network and fuse them to obtain the trajectory sequence feature representation, and then perform embedded expression to obtain the trajectory sequence feature embedding; Step S3: embedding the trajectory sequence features as input, extracting low-frequency information through the first convolutional neural network, and extracting high-frequency information through the second convolutional neural network; Step S4: inputting the low-frequency information and the high-frequency information into the frequency feature extraction module respectively to extract the low-frequency key information and the high-frequency key information; wherein the frequency feature extraction module includes three independent Mamba blocks; Step S5: converting the low-frequency key information into a low-frequency signal, and converting the high-frequency key information into a high-frequency signal; Step S6: using the low-frequency signal and the high-frequency signal as input, obtaining the time-domain trajectory sequence of the trajectory sequence data through inverse wavelet transform, and intercepting the last point of the time-domain trajectory sequence as the predicted trajectory point; Step S7: Based on the trajectory sequence data and the predicted trajectory points, iteratively calculate new predicted trajectory points; connect all the predicted trajectory points in sequence as the result of flight trajectory prediction.

2. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 1, characterized in that: Before step S1, the method further includes: A trajectory data set is obtained, and missing and duplicate trajectory data are deleted; wherein the trajectory data set includes trajectory sequence data of multiple flights.

3. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 1, characterized in that: After step S1, the following steps are also included: Unify the time granularity of trajectory sequence data through piecewise linear interpolation and resampling; The trajectory sequence data of different lengths are divided into slices of the same size X through the sliding window operation 1:t ; where t is the slice length.

4. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 3, characterized in that: Step S2 specifically includes: Transpose each slice X1:t and input it into the fully connected layer and ReLU activation layer in turn to obtain the corresponding time pattern D. D=ReLu(Linear(X 1：t ) T ); In the formula, Linear(·) represents the fully connected operation; Transpose the time pattern D and slice it with X 1：t Add together to obtain the trajectory sequence feature representation D′, D′=D T +X 1：t ; The trajectory sequence feature representation D′ is embedded by the Embedding method. E = Embedding(D′); Where E is the trajectory sequence feature embedding.

5. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 4, characterized in that: Step S3 specifically includes: AND low =Conv L (AND); AND high =Conv H (AND); In the formula, Conv L (·) and Conv H (·) represent the first convolutional neural network and the second convolutional neural network respectively, E low and E high They are low-frequency information and high-frequency information respectively.

6. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 5, characterized in that: Step S4 specifically includes: The low frequency information E low Input three independent Mamba blocks respectively, and extract low-frequency key information A through the attention mechanism low , Q low =Mamba0(E low ); K low =Mamba1(E low ); V low =Mamba2(E low ); A low =Q low ⊙K low ⊙V low ; In the formula, Q low , K low and V low They are all intermediate calculations; Mamba0(·), Mamba1(·) and Mamba2(·) represent three independent Mamba blocks, and ⊙ represents Hadamard multiplication; The high frequency information E high Input three independent Mamba blocks respectively, and extract high-frequency key information A through the attention mechanism high , <h2 style=";text-align:left;direction:ltr">Q<h2 style=";text-align:left;direction:ltr"> high <h2 style=";text-align:left;direction:ltr"> =Mamba3(E<h2 style=";text-align:left;direction:ltr"> high <h2 style=";text-align:left;direction:ltr"> ); K high =Mamba4(E high ); V high =Mamba5(E high ); A high =Q high ⊙K high ⊙V high ; In the formula, Q high , K high and V high All of them are intermediate calculations; Mamba3(·), Mamba4(·) and Mamba5(·) represent three independent Mamba blocks respectively.

7. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 6, characterized in that: Step S4 also includes: Mamba0(·) and Mamba3(·) are the same Mamba block.

8. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 6, characterized in that: Step S5 specifically includes: For low frequency key information A low Convolution operation and linear processing are performed in sequence to obtain the low-frequency signal F low , F low =Linear(Conv l (A low )); In the formula, Conv l (·) indicates the third convolutional layer; For high frequency key information A high Convolution operation and linear processing are performed in sequence to obtain the high-frequency signal F high , F high =Linear(Conv h (A high )); In the formula, Conv h (·) indicates the fourth convolutional layer.

9. A multi-channel-frequency based flight trajectory prediction method as claimed in claim 8, characterized in that: Step S6 specifically includes: Y 1:t+1 =IDWT(F low ,F high ) Where Y 1:t+1 represents the time domain trajectory sequence, IDWT(·) represents the inverse wavelet transform; Intercept the last point Y t+1 is the trajectory prediction point P.

Images