Short-term flight path prediction method based on aircraft intention and data driving

Abstract

The invention discloses a short-term flight path prediction method based on aircraft intention and data driving. The method comprises the following steps: firstly, selecting an airport to obtain a large amount of track ADS-B feature information, and preprocessing data by using a DBSCAN and a Newton interpolation method; secondly, weight distribution is carried out on the track based on segmented track density, and a segmented nominal track is obtained; thirdly, based on a TBO four-dimensional track prediction strategy, an aircraft intention model used for short-term prediction is constructed; and finally, inputting intention information and the processed ADS-B position information to train a Seq2Seq model, and obtaining a short-term prediction result of different synchronization lengths. According to the method, the nominal track is obtained in a segmented manner, so that the characteristics of different legs can be accurately distinguished from the macroscopic level; the nominal track is obtained based on the track density, so that the main development trend of the track can be sensed from the microscopic level; the intention information of the aircraft arriving at the aircraft is obtained from the time and air aspects, and the data driving model is fused, so that the accuracy of short-term prediction can be enhanced.

Description

Technical Field

[0001] The invention relates to a short-term track prediction method based on aircraft intention and data driving, and belongs to the field of track prediction. Background Art

[0002] The problem of trajectory prediction involves multiple aspects such as meteorology, performance, and intention. Currently, the methods used for trajectory prediction are mainly divided into three categories, namely state estimation models, dynamic models, and machine learning models. However, most of them are used in the strategic and pre-tactical stages to implement long-term predictions, and there are fewer short-term predictions. In addition, the prediction accuracy needs to be improved. Short-term trajectory prediction belongs to the tactical stage. The prediction accuracy greatly affects the probability of conflict and the accuracy of conflict detection, and is crucial to traffic management, arrival management, anomaly detection, etc. Therefore, the problem of short-term online trajectory prediction needs to be solved urgently. Summary of the invention

[0003] The present invention proposes a short-term trajectory prediction method based on aircraft intention and data-driven, which can achieve high-precision prediction in a short time and can be used for conflict detection and resolution. The steps are as follows:

[0004] Step 1: Select the ADS-B data of the flight paths between two airports within one year, and use DBSCAN and Newton interpolation method to process the data, including filling in missing data and removing abnormal data.

[0005] Step 2: Extract nominal tracks in segments. Different tracks are assigned weights based on the density distribution method, and the nominal track of the current segment is obtained by weighted summation.

[0006] Step 3: Based on the TBO four-dimensional track prediction strategy, the time changes caused by angle changes and arc distance differences are introduced into the track intention to enrich the short-term track intention information.

[0007] Step 4: Input location information and intent information of different step lengths, train the Seq2Seq model with a built-in neural network, and obtain output content of different step lengths.

[0008] Step 1 specifically includes:

[0009] Step 1.1: Perform MinMax normalization on the longitude, latitude, heading, and speed in the ADS-B data according to the following formula to make the track characteristics dimensionless and speed up the subsequent model training.

[0010]

[0011] Among them, x i ' is the standard value, x i is the original value, x min is the minimum characteristic value, x max is the maximum value of the feature;

[0012] Step 1.2: Use the DBSCAN method based on density clustering to cluster the track positions, determine the neighborhood radius Eps and the threshold MinPts for defining the core point, and automatically obtain the optimal number of clusters and eliminate abnormal tracks.

[0013] Step 1.3: Take T as the time slice and reconstruct the eigenvalue corresponding to the timestamp through the mean. Given the timestamp t 1 , t 2 , feature x 1 , x 2 If t 2 -t 1 <T, then the reconstructed standard eigenvalue is

[0014]

[0015] Step 1.4: Since a single track has no periodicity and no specific rules, the Newton interpolation method can be used to fill in the missing values. Use the known track points to establish a suitable interpolation function f(x),

[0016]

[0017] Among them, f[x 0 ,…,x n ] is the n-order difference quotient, which is calculated as follows:

[0018]

[0019] Missing point p i The function value f(p) can be obtained by interpolating the function f(x) i ), use the obtained approximate value as the missing value to complete the missing trajectory features.

[0020] Step 2 specifically includes:

[0021] Step 2.1: Divide the flight into m segments according to the characteristics of different flight segments M = {M 1 ,M 2 ,…,M m}, for the current flight segment M i , give larger weights to tracks with larger density, and smaller weights to tracks with smaller density. Take the outermost and innermost tracks as the selection range, and when calculating the density of each track, take the track as the center, expand the selection range to both sides, and calculate the total tracks a within the range ij , the line density is:

[0022]

[0023] Step 2.2: Perform z-score standardization on the line density.

[0024]

[0025] where μ i is the mean of the track set of the ith flight segment, σ i is the variance.

[0026] Step 2.3: Get the flight segment weight set ω i ={ω i1 ,ω i2 ,…,ω in}(i=1,2,…,m), the nominal track of each segment is obtained by density-weighted summation, M i The nominal track of the segment is:

[0027]

[0028] Among them l j is the jth track.

[0029] Step 3 specifically includes:

[0030] Step 3.1: Based on the TBO four-dimensional track strategy, the angle difference and arc distance are introduced into the track intention to construct the short-term track intention vector.

[0031] P 1 is the current nominal track point, P 2 is the next nominal track point, P 0 For the current identified track point,

[0032] d 0 P 1 is the center of the circle and the radius is l 0 The arc distance of the concentric circle with l, α 0 for l 0 and the angle between l.

[0033] Step 3.2: Identify the current position, use the Euclidean distance to obtain the nearest nominal track point, and dynamically adjust the P 1 The radius of the circle with the center being r, P 0 Contained in the circle, but not in the circle with P 2 is a circle with center at and radius r.

[0034] Step 3.3: α 0 is for 0 The angular difference between l and l is used to describe the directional intention of the aircraft.

[0035] α 0 = l 0 -l

[0036] If l 0 Before l, α0 Take positive value, l 0 If it lags behind l, it takes a negative value.

[0037] d 0 It represents the controllable distance from the nominal track point to the current track point and the next track point. Taking the quotient of the current ground speed, we get T 0 , which can control the estimated time to reach the next nominal track point,

[0038]

[0039] If l>l 0 , then T 0 Take positive value, otherwise take negative value.

[0040] Therefore, the intention information I = [α 0 ,T 0 ].

[0041] Step 4 specifically includes:

[0042] Step 4.1: The present invention uses the Seq2Seq model for prediction, and its encoding network and decoding network can be set to LSTM, GRU and other structures. The input information is X t ={x t-q ,x t-q+1 ,…,x t}, length is q, x i =[t i ,lon i ,lat i ,α 0i ,T 0i ], the output information is Y t ={y t+1 ,y t+2 ,…,y t+p}, length p, y j =[t j ,lon i ,lat i ]. The track feature information is divided into position information P = [t, lon, lat] and intention information I = [α 0 ,T 0 ].

[0043] Step 4.2: Seq2Seq for input information X t Encode and process the input information through the following formula.

[0044]

[0045] in, Learn parameters for the encoder.

[0046] According to the above content, the hidden sequence H can be obtained t ={h t-q ,h t-q+1 ,…,h t}.

[0047] Step 4.3: The decoder converts the encoder output Z j =q(h 1 ,h 2 ,…,h q ), the predicted value at the previous moment and the potential representation u j-1 As the input of the decoder, after the decoder, the parameters in the network are learned, and the predicted value of the next p moments can be obtained

[0048] Step 4.4: Use the root mean square error (RMSE), mean absolute error (MAE), and mean relative error (MRE) calculation methods to train the model so that the interpolation between the predicted value and the true value is minimized.

[0049]

[0050]

[0051]

[0052] Among them, p i is the true value, is the predicted value.

[0053] The beneficial effects achieved by the present invention are as follows: obtaining the nominal track in segments can more accurately distinguish the characteristics of different segments from a macro level, and obtaining the nominal track based on the track density can perceive the main development trend of the track from a micro level; adding aircraft intention information that can control the arrival of the aircraft from both time and space aspects can enhance the short-term accuracy of the prediction; because different subjects have different requirements for short-term tracks, it is necessary to output prediction information of different time scales, and the Seq2Seq structure can meet this requirement. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of the method of the present invention;

[0055] Figure 2 This is a schematic diagram of the track intention in the normal operation mode;

[0056] Figure 3 It is a schematic diagram of the Seq2Seq structure;

[0057] Figure 4 ADS-B data diagram;

[0058] Figure 5Schematic diagram of ZUUU-ZSHC track. Specific embodiments

[0059] In order to make the purpose, technical solution and key points of the present invention clearer, the flight path between Chengdu Shuangliu Airport (ZUUU) and Hangzhou Xiaoshan Airport (ZSHC) is selected to further illustrate the present invention in combination with the accompanying drawings.

[0060] The present invention proposes a short-term trajectory prediction method based on aircraft intention and data-driven, the flow chart is as follows Figure 1 The specific implementation steps are as follows:

[0061] Step 1: Use the Python crawler variflight to obtain the ADS-B data of flight 3U8088 from ZUUU to ZSHC from July 1, 2022 to June 30, 2023, such as Figure 4 DBSCAN and Newton interpolation method are used for data processing, including filling in missing data and removing abnormal data.

[0062] Step 1.1: Perform MinMax normalization on the longitude, latitude, heading, and speed in the ADS-B data according to the following formula to make the track characteristics dimensionless and speed up the subsequent model training.

[0063]

[0064] Step 1.2: Use the DBSCAN method based on density clustering to cluster the track positions, set the neighborhood radius Eps = 0.1, define the core point threshold MinPts = 300, obtain the best number of clusters as 2, and remove abnormal tracks, such as Figure 5 shown.

[0065] Step 1.3: In this example, 10 minutes is used as the time slice, and the eigenvalue corresponding to the timestamp is reconstructed by the mean. 1 , t 2 , feature x 1 , x 2 If t 2 -t 1 <10min, then the reconstructed standard eigenvalue is

[0066]

[0067] Step 1.4: Since a single track has no periodicity and no specific rules, the Newton interpolation method is used to fill in the missing values. Use the known track points to establish a suitable interpolation function f(x),

[0068]

[0069] Among them, f[x 0 ,…,x n ] is the n-order difference quotient, which is calculated as follows:

[0070]

[0071] Missing point p i The function value f(p) can be obtained by interpolating the function f(x) i ), use the obtained approximate value as the missing value to complete the missing trajectory features.

[0072] Step 2: Extract nominal tracks in segments. Different tracks are assigned weights based on the density distribution method, and the nominal track of the current segment is obtained by weighted summation.

[0073] Step 2.1: Divide the flight into m segments according to the characteristics of different flight segments M = {M 1 ,M 2 ,…,M m}, for the current flight segment M i , give larger weights to tracks with larger density, and smaller weights to tracks with smaller density. Take the outermost and innermost tracks as the selection range, and when calculating the density of each track, take the track as the center, expand the selection range to both sides, and calculate the total tracks a within the range ij , the line density is:

[0074]

[0075] Step 2.2: Perform z-score standardization on the line density.

[0076]

[0077] where μ i is the mean of the track set of the ith flight segment, σ i is the variance.

[0078] Step 2.3: Get the flight segment weight set ω i ={ω i1 ,ω i2 ,…,ω in}(i=1,2,…,m), the nominal track of each segment is obtained by density-weighted summation, M i The nominal track of the segment is:

[0079]

[0080] Among them l j is the jth track.

[0081] Step 3: Based on the TBO four-dimensional track prediction strategy, the time changes caused by angle changes and arc distance differences are introduced into the track intention to enrich the short-term track intention information.

[0082] Step 3.1: Based on the TBO four-dimensional track strategy, the angle difference and arc distance are introduced into the track intention to construct a short-term track intention vector. The schematic diagram is as follows: Figure 2 shown.

[0083] P 1 is the current nominal track point, P 2 is the next nominal track point, P 0 For the current identified track point,

[0084] d 0 P 1 is the center of the circle and the radius is l 0 The arc distance of the concentric circle with l, α 0 for l 0 and the angle between l.

[0085] Step 3.2: Identify the current position, use the Euclidean distance to obtain the nearest nominal track point, and dynamically adjust the P 1 The radius of the circle with the center being r, P 0 Contained in the circle, but not in the circle with P 2 is a circle with center at and radius r.

[0086] Step 3.3: α 0 is for 0 The angular difference between l and l is used to describe the directional intention of the aircraft.

[0087] α 0 = l 0 -l

[0088] If l 0 Before l, α 0 Take positive value, l 0 If it lags behind l, it takes a negative value.

[0089] d 0 It represents the controllable distance from the nominal track point to the current track point and the next track point. Taking the quotient of the current ground speed, we get T 0 , which controls the estimated time to reach the next nominal track point.

[0090]

[0091] If l>l 0 , then T 0 Take positive value, otherwise take negative value.

[0092] Therefore, the intention information I = [α 0 ,T 0].

[0093] Step 4: Input location information and intent information of different step lengths. Here, the Seq2Seq model with a built-in LSTM network is trained to obtain output content of different step lengths.

[0094] Step 4.1: The present invention uses the Seq2Seq model to make predictions, such as Figure 3 As shown in Figure 1, the encoding network and decoding network are set to LSTM structure. The input information is X t ={x t-q ,x t-q+1 ,…,x t}, length is q, x i =[t i ,lon i ,lat i ,α 0i ,T 0i ], the output information is Y t ={y t+1 ,y t+2 ,…,y t+p}, length p, y j =[t j ,lon i ,lat i ]. The track feature information is divided into position information P = [t, lon, lat] and intention information I = [α 0 ,T 0 ].

[0095] Step 4.2: Seq2Seq for input information X t Encode and process the input information through the following formula.

[0096]

[0097] in, Learn parameters for the encoder.

[0098] According to the above content, the hidden sequence H can be obtained t ={h t-q ,h t-q+1 ,…,h t}.

[0099] Step 4.3: The decoder converts the encoder output Z j =q(h 1 ,h 2 ,…,h q ), the predicted value at the previous moment and the potential representation u j-1As the input of the decoder, after the decoder, the parameters in the network are learned, and the predicted value of the next p moments can be obtained

[0100] Step 4.4: Use the root mean square error calculation method to train the model so that the interpolation between the predicted value and the true value is minimized.

[0101]

[0102] Among them, p i is the true value, is the predicted value.

[0103] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A short-term trajectory prediction method based on aircraft intention and data-driven, characterized by: The following steps are involved: Step 1: Select the ADS-B data of the flight paths between two airports within one year, and use DBSCAN and Newton interpolation method to process the data, including filling in missing data and removing abnormal data; Step 2: Extract the nominal track in sections, assign weights to different tracks based on the density distribution weight method, and obtain the nominal track of the current section by weighted summation; Step 3: Based on the TBO four-dimensional track prediction strategy, the time changes caused by angle changes and arc distance differences are introduced into the track intention to enrich the short-term track intention information; Step 4: Input location information and intent information of different step lengths, train the Seq2Seq model with a built-in neural network, and obtain output content of different step lengths.

2. The method for short-term trajectory prediction based on aircraft intention and data-driven according to claim 1, characterized in that: Step 1 specifically includes: Select the route between two airports, import the ADS-B data of the route within this range and perform pre-processing; Step 1.1: Perform MinMax normalization on the longitude, latitude, heading, and speed in the ADS-B data according to the following formula to make the track characteristics dimensionless and speed up the subsequent model training; Among them, x i ' is the standard value, x i is the original value, x min is the minimum characteristic value, x max is the maximum value of the feature; Step 1.2: Cluster the track positions using the DBSCAN method based on density clustering, determine the neighborhood radius Eps and the threshold MinPts for defining the core point, automatically obtain the optimal number of clusters, and eliminate abnormal tracks; Step 1.3: Take T as the time slice and reconstruct the eigenvalue corresponding to the timestamp by the mean; if the timestamps t1, t2, features x1, x2 are known and t2-t1<T, then the reconstructed standard eigenvalue is Step 1.4: Since a single track has no periodicity and no specific rules, the Newton interpolation method is used to fill in the missing values; the known track points are used to establish a suitable interpolation function f(x), Among them, f[x0,…,x n ] is the n-order difference quotient, which is calculated as follows: Missing point p i The function value f(p) can be obtained by interpolating the function f(x) i ), use the obtained approximate value as the missing value to complete the missing trajectory features.

3. The method for short-term trajectory prediction based on aircraft intention and data-driven according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1: Divide the flight into m segments according to the characteristics of different segments M = {M1, M2, …, M m }, for the current flight segment M i , giving larger weights to tracks with larger line density and smaller weights to tracks with smaller line density; taking the outermost and innermost tracks as the frame selection range, when calculating the density of each track, take the track as the center, expand the frame selection range to both sides, and calculate the total tracks a within the range ij , the line density is: Step 2.2: Perform z-score standardization on the line density. where μ i is the mean of the track set of the ith flight segment, σ i is the variance; Step 2.3: Get the flight segment weight set ω i ={ω i1 ,ω i2 ,…,ω in }(i=1,2,…,m), the nominal track of each segment is obtained by density-weighted summation, M i The nominal track of the segment is: Among them l j is the jth track.

4. The method for short-term trajectory prediction based on aircraft intention and data-driven according to claim 1, characterized in that: Step 3 specifically includes: Step 3.1: Based on the TBO four-dimensional track prediction strategy, the angle difference and arc distance are introduced into the track intention to construct the short-term track intention vector; P1 is the current nominal track point, P2 is the next nominal track point, and P0 is the current identified track point. d0 is the arc distance of the concentric circles with P1 as the center and radius l0 and l, and α0 is the angle between l0 and l; Step 3.2: Identify the current position, use the Euclidean distance to obtain the nearest nominal track point, and dynamically adjust the radius r of the circle with P1 as the center to include P0 in the circle, but not in the circle with P2 as the center and radius r; Step 3.3: α0 is the angular difference between l0 and l, which is used to describe the directional intention of the aircraft. α0＝l0-l If l0 precedes l, α0 takes a positive value, and if l0 lags behind l, it takes a negative value; d0 represents the controllable distance from the nominal track point to the current track point and to the next track point; by taking the quotient of the current ground speed, we get T0, which can control the estimated time to reach the next nominal track point. If l>l0, T0 takes a positive value, otherwise it takes a negative value; Therefore, the intention information I = [α0, T0] is constructed.

5. The method for short-term trajectory prediction based on aircraft intention and data-driven according to claim 1, characterized in that: Step 4 specifically includes: Step 4.1: Use the Seq2Seq model for prediction. The encoding network and decoding network can be set to LSTM and GRU structures. The input information is X t ={x t-q ,x t-q+1 ,…,x t }, length is q, x i =[t i ,lon i ,lat i ,α 0i ,T 0i ], the output information is Y t ={y t+1 ,y t+2 ,…,y t+p }, length p, y j =[t j ,lon i ,lat i ]; the track characteristic information is divided into position information P = [t, lon, lat] and intention information I = [α0, T0]; Step 4.2: Seq2Seq for input information X t Encode and process the input information through the following formula; in, Learn parameters for the encoder; According to the above content, the hidden sequence H is obtained t ={h t-q ,h t-q+1 ,…,h t }; Step 4.3: The decoder converts the encoder output Z j =q(h1,h2,…,h q ), the predicted value at the previous moment and the potential representation u j-1 As the input of the decoder, after the decoder, the parameters in the network are learned to obtain the predicted value of the next p moments Step 4.4: Use the root mean square error RMSE, mean absolute error MAE, and mean relative error MRE calculation methods to train the model so that the interpolation between the predicted value and the true value is minimized; Among them, p i is the true value, is the predicted value.

Images