Method for Mining Aircraft Flight Modes in Airport Terminal Area Based on Standard Flight Procedures

Abstract

The present invention discloses a method for mining the flight patterns of aircraft in the airport terminal area based on standard flight procedures. First, quality analysis is performed on the arrival and departure data of the aircraft and the standard flight procedures, and the required feature vectors are obtained through preprocessing. Secondly, a method based on standard procedure guidance is used to mine the flight patterns of the flight tracks. In the process of determining the central track during clustering, the standard track is used as part of the central clustering track, and the distance calculation of the tracks is realized by using the method of calculating the similarity measure through dynamic time warping. The present invention improves the selection method of the central track on the basis of the original k-means mining flight pattern, classifies the tracks with the standard procedure as the main clustering center, reduces the number of iterations, improves the clustering accuracy, and makes the result more interpretable.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil aviation air traffic control automation and intelligence, and particularly relates to a method for mining aircraft flight patterns in an airport terminal area based on standard flight procedures. Background Technique

[0002] With the rapid development of civil aviation, the air traffic flow has increased sharply, resulting in a more complex air traffic situation than before. While the arrival and departure flight procedures cannot meet the current airspace status, how to mine the mainstream flight patterns from a large amount of track data has become a problem to be solved today. For the airport terminal area, analyzing track data and identifying flight patterns in the airport terminal area are of crucial importance to airlines and air traffic control. For airlines, the prevailing flight patterns of aircraft can intuitively reflect the overall operation status of flights, so that flights can be adjusted. Air traffic controllers can more conveniently judge track anomalies according to flight patterns, and at the same time can use relevant data to optimize flight procedures.

[0003] Currently, the mainstream method for mining flight patterns is the clustering-based method. Clustering is a widely used unsupervised and semi-supervised technique that groups similar data instances into clusters according to the definition of pairwise distances or similarity functions. It has the advantages of simple structure, reliable algorithm, high computational efficiency, and applicability to large data sets, so it is widely used to identify relevant flight operation events. By clustering the tracks, the central tracks representing the mainstream traffic patterns can be extracted from the massive track data, so as to obtain the distribution status of the traffic flow in the terminal area.

[0004] Currently, there are three problems to be solved in using the clustering method to mine flight patterns: First, the number of clusters in track clustering is crucial for the clustering accuracy. With the continuous increase in the number of tracks, the distribution of air traffic flow becomes more complex, and it will be more difficult to determine the number of clusters. Second, the current large amount of track data will greatly increase the number of clustering iterations, so a suitable method needs to be found to improve the clustering efficiency and accuracy. At the same time, since the currently obtained track data is composed of track points, the mismatch of the time series of planned track points will affect the clustering accuracy.

[0005] Flight mode mining is a basic technology for four-dimensional trajectory prediction technology, arrival and departure sequencing technology, and flight simulation technology. The present invention proposes a flight mode mining and clustering method based on standard flight procedures. First, the obtained trajectory data is analyzed to obtain the trajectory point coordinates and trajectory point data such as climb rate, speed, heading, and angle as the feature data for clustering. During the clustering process, the standard flight procedure provides a solution for the selection of the central trajectory, thereby improving the accuracy of flight mode mining. This method determines the clustering center of the initial flight mode through the standard flight procedure, and on this basis, uses the dynamic time warping method to achieve the matching and classification of trajectory points. During the iterative process, each iteration result is associated with the standard flight procedure, thereby realizing the mining and classification of prevalent flight modes related to the standard procedure. The present invention solves the problems of difficult matching of trajectory point time series and determination of the number of clustering clusters, and improves the mining accuracy of flight modes. Summary of the Invention

[0006] Purpose of the Invention: In order to use the standard flight procedure to reduce the number of clustering iterations and improve the accuracy of the clustering results, the present invention proposes an aircraft flight mode mining method in the airport terminal area based on the standard flight procedure; using the standard flight procedure trajectory as part of the initial central trajectory, and always maintaining the standard flight procedure trajectory as the center of each flight mode during each iteration process, so that the mined flight mode is more in line with the actual situation.

[0007] Technical Solution: The present invention provides an aircraft flight mode mining method in the airport terminal area based on the standard flight procedure, specifically including the following steps:

[0008] Step 1: Obtain takeoff trajectories, landing trajectories, overflight trajectories data and standard flight procedures for arrival and departure;

[0009] Step 2: Conduct quality analysis on the obtained data and perform preprocessing;

[0010] Step 3: Perform coordinate transformation on the trajectory point data and construct a multi-dimensional trajectory feature vector;

[0011] Step 4: Determine the central trajectory of each flight mode and obtain the corresponding flight mode features.

[0012] Further, the takeoff trajectories, landing trajectories, and overflight trajectories data in Step 1 include information recording time, flight number, aircraft location information, heading, climb rate, and aircraft speed.

[0013] Further, Step 2 includes the following steps:

[0014] Step 2.1: Aircraft trajectory merging: The obtained trajectory data is classified by hour, and the trajectory point information within each hour is a group;

[0015] Step 2.2: Number the flight tracks: When the time interval between adjacent flight track points with the same flight number exceeds 15 minutes, it is defined that these two flight track points come from different flights; the flight tracks of each day are numbered and differentiated by combining the year, month, day and the track ordinal number.

[0016] Step 2.3: For duplicate values, missing values and outliers: For duplicate values, the method of deletion is adopted, keeping the first set of duplicate data and deleting the rest; for missing values, interpolation or deletion methods are used for processing; for data with a too small missing rate, the regression method is used to fill the missing data, and for data with a too large missing rate, all information of the corresponding flight track point is deleted; for outliers, the deletion method is uniformly adopted; for flight tracks with too few track points and too large interval between two adjacent points, they are deleted.

[0017] Further, the said Step 3 includes the following steps:

[0018] Step 3.1: Perform coordinate transformation on the position information of flight track points:

[0019] Convert the longitude data and latitude data returned by the secondary radar into the northeast - sky coordinates, which can reflect the distance relationship with the airport and the actual track position information of the aircraft; taking the navigation station of the airport as the coordinate origin, a rectangular coordinate system is created to obtain the e, n, u coordinate values of the northeast - sky coordinates; taking the user's coordinate origin (x0, y0, z0) as the coordinate axis origin, the position of the coordinate point (x, y, z) in the northeast - sky coordinate system is set as (e, n, u); at the same time, assuming the longitude - latitude - altitude coordinate point of the coordinate origin is LLA0=(lon0, lat0, alt0), the corresponding conversion formula is:

[0020]

[0021]

[0022] Step 3.2: Construct a multi - dimensional flight track feature vector:

[0023] The obtained flight track data includes multi - dimensional features such as longitude, latitude, altitude, speed, and heading. The shape of the flight track is completely described by constructing the climb rate a and the angle θ.

[0024] The corresponding calculation methods for constructing the two features of the climb rate a and the angle θ are as follows:

[0025]

[0026]

[0027] where, v jDenote the aircraft speed at the point at time \(t = j\); dist refers to the Euclidean distance calculated between two points; \(P_x\) j , \(P_y\) j respectively represent the horizontal and vertical coordinates corresponding to this point;

[0028] Based on the converted coordinates \(e\), \(n\), \(u\), the climb rate \(a\) and the angle \(\theta\), as well as the known information such as the speed \(v\) and the course \(\gamma\) and other features, obtain the feature data \(P\) of each trajectory point j :

[0029] \(P\) j = \((e,n,u,a,v,\gamma,\theta)\) (5)

[0030] where, \(P\) j is the trajectory point corresponding to the trajectory at time \(t = j\). Formula (5) represents the \(e\), \(n\), \(u\) coordinates of the multi-dimensional trajectory point corresponding to the trajectory at time \(t = j\), and the seven feature parameters of the climb rate, speed, course, and angle;

[0031] The trajectory \(TR\) is a set of trajectory point sequences. \(j\in[1,m]\) is the trajectory point number, and \(m\) is the total number of trajectory points. Then the set of trajectory points changing with time is:

[0032] \(TR=\{P_1,P_2,\cdots,P\) j ,\cdots,P\) m \}\) (6)

[0033] Suppose there are a total of \(h\) trajectories, \(TD\) is the set of aircraft trajectories, \(i\in[1,h]\) is the track number, and \(h\) is the total number of trajectories; the trajectory set of the aircraft can be represented in the form of the following set:

[0034] \(TD = \{TR_1,TR_2,\cdots,TR\) i ,\cdots,TR\) h \}\). (7)

[0035] Furthermore, step 4 includes the following steps:

[0036] Step 4.1: Search for the track closest to the standard flight procedure from the historical track data:

[0037] Find the actual track closest to the standard flight procedure: From all actual tracks \(TD\) B = \{TR b1 ,TR b2 ,\cdots,TR bQ \}, use the Euclidean distance method to calculate the distance from all standard procedures \(TD\) A = \{TR a1 ,TR a2 ,\cdots,TR aR \}, and from \(TD\) BFind the actual trajectory closest to each standard program trajectory in it, and use TD cen to represent the set of these trajectories;

[0038] Step 4.2: Match the actual trajectory with TD cen as follows:

[0039] Calculate the dynamic time warping distances between each of the other trajectories in the set TD B and these TD cen trajectories respectively:

[0040] δ(P ay , P bz ) = (P ay - P bz ) 2 (8)

[0041]

[0042] where P ay is defined as the central trajectory sequence of TR a = {P a1 , P a2 ,..., P ay ,..., P aY}, P bz is defined as the actual track sequence of TR b = {P b1 , P b2 ,..., P bz ,..., P bZ}, d(P ay , P bz ) is the DTW distance between two points, and the sum of the DTW distances of the track points is used as the correlation degree of this group of tracks; Select several actual tracks near each track in the TD cen set as the matching tracks of its corresponding cluster in TD C ∈ TD B to complete the initial clustering of the trajectory based on the standard flight procedure; The actual trajectory TD B is divided into three types of trajectories: the cluster trajectory TD cen corresponding to each trajectory in TD cen , the remaining trajectory TD C , and the remaining trajectory TD D ; Select a new clustering center trajectory from the remaining trajectories;

[0043] Step 4.3: Randomly select the remaining initial center trajectory:

[0044] From the remaining trajectories TD B in TD DRandomly select k - r trajectories from them as the remaining initial central trajectories, denoted as TD’ cen , which is TD D Match the other trajectories in; the set TD cen and the set TD’ cen constitute all the initial central trajectories; the remaining trajectories are assigned to the clusters corresponding to the trajectories in the TD’ cen set;

[0045] Step 4.4: Matching degree analysis: According to the k central trajectories, calculate the DTW distance d(P ay , P bz ) between the actual trajectory and each central trajectory; store the sub - result calculated by d(P ay , P bz ) in the cumulative distance matrix D(Y, Z), where D(Y, Z) is a matrix with Y rows and Z columns, which is convenient for storing all the results of d(P ay , P bz ). d(Y, Z) is used as the optimal global distance, representing the similarity between two trajectories;

[0046] Calculate the similarity distance between the l - th central trajectory sequence TR l and the given actual track sequence, denoted as d l (Y, Z); introduce w for matching degree analysis:

[0047] w = min(d l (X, Y)) (10)

[0048] Each actual trajectory TR s corresponds to the matching degree w s of the matched central trajectory; adopt the min - max normalization method to linearly map the data into the [0, 1] interval:

[0049]

[0050] where, min(w s ) and max(w s ) correspond to the minimum and maximum values of the w values obtained from all actual trajectories respectively. The finally obtained w’ s represents the normalized matching degree of the s - th actual trajectory TR s ;

[0051] Step 4.5: Updated clustering center: For the class where the central trajectory is TD cen is located, the coordinates of each point of the central trajectory do not change. The central trajectory is TD’ cenThe class where it is located calculates the average value of each class of coordinates, selects the trajectory closest to it as the new clustering center, and updates the original TD'. cen The coordinate positions corresponding to the trajectories, and thus the first iteration is completed;

[0052] Step 4.6: Repeat Step 4.4 and Step 4.5 until the clustering center no longer changes or the number of clustering times reaches the requirement, that is, output the result after the clustering result reaches a stable state.

[0053] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows:

[0054] 1. The present invention uses a standard flight procedure to assist in obtaining the initial center trajectory, making the separation of the flight tracks associated with the standard flight procedure stipulated by air traffic control; clustering using this method can obtain a clustering result corresponding to the actual flight mode, which is more in line with the actual flight law of the aircraft;

[0055] 2. Using the standard flight procedure trajectory to replace the randomly selected k - means initial center trajectory, this improvement can greatly reduce the number of clustering iterations in the case of complex and dense flight track distributions, and at the same time can reduce the steps of finding a new clustering center in each iteration, improving the calculation efficiency;

[0056] 3. While clustering using spatial features, feature data such as the speed, heading, and climb rate of the aircraft are introduced, enabling the flight tracks to be distinguished according to different speeds, different climb rates, etc. While maximizing the utilization of historical flight track data information, the clustering result is more interpretable. Description of the Drawings

[0057] Figure 1 is a flow chart of the present invention;

[0058] Figure 2 is a schematic diagram of angle calculation;

[0059] Figure 3 is a schematic diagram of DTW path matching principle;

[0060] Figure 4 is a distribution diagram of flight tracks after clustering for approach (left) and departure (right);

[0061] Figure 5 is a distribution diagram of clustering centers for approach (left) and departure (right). Specific Embodiments

[0062] The following further describes the present invention in detail with reference to the drawings.

[0063] The present invention provides a method for mining the flight mode of aircraft in the airport terminal area based on a standard flight procedure, as Figure 1 shown, including the following steps:

[0064] Step 1: Obtain takeoff and landing track data, overflight track data, and standard flight procedures.

[0065] Based on the track data of a certain area, including takeoff tracks, landing tracks, and overflight tracks, the track information within the specific terminal area of the airport is obtained. The secondary radar data selected for the experiment needs to include all takeoff and landing tracks within the airport terminal area, that is, the track data within a radius of 80 kilometers centered on the airport, including information recording time, flight number, aircraft position information (longitude, latitude, and altitude), heading, rate of climb and descent, aircraft speed, etc. All flight routes are divided into approach tracks and departure tracks according to the takeoff and landing altitudes. The approach and departure tracks will be clustered separately in subsequent operations.

[0066] Step 2: Conduct quality analysis on the obtained data and perform preprocessing.

[0067] Step 2.1: Merge aircraft tracks.

[0068] The obtained track data is classified by hour, and the track point information within each hour is grouped. In order to ensure that all track points of a single track are concentrated in the same group, the track data for a day is merged, and the track data for 24 hours is grouped into one group.

[0069] Step 2.2: Assign numbers to the tracks.

[0070] To ensure the distinction between flights, the operating intervals of flights with the same flight number will exceed the operating time of a single flight. Therefore, when the time interval between adjacent track points of the same flight number exceeds 15 minutes, it can be determined that these two track points come from different flights. To avoid the situation where the same flight number may occur on different days, the tracks of each day's flights can be numbered and differentiated using the method of adding the track ordinal number to the year, month, and day of the flight, so that each track within a year has a unique number.

[0071] Step 2.3: Duplicate values, missing values, and outliers.

[0072] Compare and analyze all the attributes of the track points. For the cases where the attributes are exactly the same or there are null values in the attributes, data needs to be made more complete through methods such as deletion and interpolation. For outliers in the data, they can be discriminated by clustering methods. The data is divided into multiple clusters, and obvious outliers are found for data correction. In the records of radar data, there may be tracks with short tracks and too few analyzable track points. Therefore, it is necessary to find out this type of track that may cause certain interference to clustering. For duplicate values, the method of deletion is adopted, keeping the first set of duplicate data and deleting the rest; for missing values, interpolation or deletion methods are used for processing. For data with a too small missing rate, the regression method is used to fill in the missing data, and for data with a too large missing rate, all the information of the track point is deleted; for outliers, the method of deletion is uniformly adopted.

[0073] Step 2.4: Tracks with too few track points and too large intervals between two adjacent points.

[0074] At the same time, for tracks with too few track points and too large intervals between two adjacent points, these will all affect the clustering results, and the corresponding tracks should be deleted.

[0075] Step 3: Perform coordinate transformation on the track point data and construct a multi-dimensional track feature vector.

[0076] Step 3.1: Coordinate transformation.

[0077] The coordinate data returned by the secondary radar is longitude data and latitude data. Converting it into the northeast-up coordinate to reflect the distance relationship with the airport can better reflect the actual track position information of the aircraft. Taking the navigation station of the airport as the coordinate origin, a rectangular coordinate system is created to obtain the e, n, u coordinate values of the northeast-up coordinate. Taking the user's coordinate origin (x0, y0, z0) as the coordinate axis origin, the position of the coordinate point (x, y, z) in the northeast-up coordinate system is set as (e, n, u). At the same time, the longitude-latitude-altitude coordinate point of the coordinate origin is set as LLA0 = (lon0, lat0, alt0). Then the corresponding conversion formula is:

[0078]

[0079]

[0080] Step 3.2: Construct the feature vector.

[0081] The track data obtained by secondary radar includes multi-dimensional features such as longitude, latitude, altitude, speed, and heading. At the same time, it is difficult to completely represent the shape features of the track only by using the three-dimensional space data of the track. The shape of the track can be more completely described by constructing the climb rate a and the angle θ. The multi-dimensional track feature vector obtained thereby can better describe different flight modes and improve the interpretability of the clustering results.

[0082] The corresponding calculation methods for constructing the two features of the climb rate a and the angle θ are as follows:

[0083]

[0084]

[0085] Among them, v j represents the aircraft speed at this point at time t = j; dist refers to the Euclidean distance calculated between two points; Px j , Py j respectively represent the horizontal and vertical coordinates corresponding to this point. Among them, the calculated angle is between 0° and 360°, Figure 2 indicates that the included angles calculated for different track points are different. And since the first point P1 and the last point P m cannot form an included angle, the angle of this track point is replaced by the angle of P2 or P m-1 .

[0086] After obtaining the transformed coordinates e, n, u, and at the same time based on the calculated climb rate a and angle θ, as well as the known information such as speed v and heading γ and other features, the characteristic data P of each track point is finally obtained j :

[0087] P j =(e, n, u, a, v, γ, θ) (5)

[0088] Among them, P j is the track point corresponding to the track at time t = j. Formula (5) represents the e, n, u coordinates of the multi-dimensional track point corresponding to the i-th track at time t = j, and the 7 characteristic parameters of the climb rate, speed, heading, and angle.

[0089] The track TR i is a set of track point sequences, j ∈ [1, m] is the track point number, and m is the total number of track points. Then the set of track points changing with time is:

[0090] TR i ={P i1 , P i2 ,..., P ij ,..., P im} (6)

[0091] Suppose there are h trajectories in total. TD is the set of aircraft trajectories. Let i ∈ [1, h] be the track number, and h be the total number of trajectories. The trajectory set of the aircraft can be represented in the form of the following set:

[0092] TD = {TR1, TR2,..., TR i ,..., TR h} (7)

[0093] Step 4: Determine the central trajectory of each type of flight mode and obtain the corresponding flight mode characteristics.

[0094] Step 4.1: Search for the track closest to the standard flight procedure from the historical trajectory data.

[0095] Find the actual trajectory closest to the standard flight procedure: From all the actual trajectories TD B = {TR b1 , TR b2 ,..., TR bQ}, calculate the distance from all the standard procedures TD A = {TR a1 , TR a2 ,..., TR aR} using the Euclidean distance method, and find the actual trajectory closest to each standard procedure trajectory from TD B . Denote the set of these trajectories by TD cen .

[0096] Step 4.2: Match the actual trajectories with TD cen .

[0097] Calculate the dynamic time warping (DTW) distances between the other trajectories in the set TR B and these TD cen trajectories respectively. The dynamic time warping method is widely used to measure the similarity of unequal-length time series. In time series, due to the problems of unequal time series lengths or displacements of different series on the time axis, the traditional Euclidean distance cannot accurately calculate the similarity between two time series. Therefore, the global similarity of the route can be obtained by effectively shortening or extending the time series. As Figure 3 shown, the two time series represent the flight procedure waypoint sequence and the actual waypoint sequence respectively. When the two waypoint sequences cannot be directly and obviously matched, the sum of the distances of the corresponding similar points is calculated through DTW to obtain the similarity distance between the two trajectories, that is, the warping path distance.

[0098] The calculation method of DTW is specifically as follows:

[0099] δ(P ay ,P bz ) = (P ay - P bz ) 2 (8)

[0100] where P ay is defined as the central trajectory sequence of TR a = {P a1 , P a2 ,..., P ay ,..., P aY}, and P bz is defined as the actual flight track sequence of TR b = {P b1 , P b2 ,..., P bz ,..., P bZ}.

[0101] To obtain the optimal sum of costs, i.e., the DTW distance, the following method can be used for recursive calculation:

[0102]

[0103] To find the cluster that each trajectory in TD cen matches, several actual trajectories are selected near each trajectory in the TD cen set as the matching trajectories of its corresponding cluster in TD C ∈ TD B , completing the initial clustering of trajectories based on standard flight procedures. So far, the actual trajectory B is divided into three types of trajectories: the cluster trajectories TD cen corresponding to each trajectory in TD cen , TD C , and the remaining trajectories TD D . A new clustering center trajectory will be selected from the remaining trajectories.

[0104] Step 4.3: Randomly select the remaining initial center trajectories.

[0105] Randomly select k - r trajectories from the remaining trajectories TD B in TD D as the remaining initial center trajectories, denoted as TD’ cen , to match the other trajectories in TR D . The set TD cen and the set TD’ cen form all the initial center trajectories. The remaining trajectories are assigned to the clusters corresponding to the trajectories in the TD’ cen set using the DTW matching principle.

[0106] Step 4.4: Matching degree analysis.

[0107] According to the k central trajectories, calculate the DTW distance d(P ay ,P bz ) between the actual trajectory and each central trajectory. Store the sub - result calculated by d(P ay ,P bz ) in the cumulative distance matrix D(Y, Z), where D(Y, Z) is a matrix with Y rows and Z columns, which is convenient for saving all the results of d(P ay ,P bz ). d(Y, Z) is used as the optimal global distance, representing the similarity between two trajectories.

[0108] Calculate the similarity distance between the l - th central trajectory sequence TR l and the given actual track sequence, denoted as d l (Y, Z); Introduce w for matching degree analysis:

[0109] w = min(d l (X, Y)) (10)

[0110] Each actual trajectory TR s corresponds to the matching degree w s of the matched central trajectory; Adopt the min - max normalization method to linearly map the data into the interval [0, 1]:

[0111]

[0112] where, min(w s ) and max(w s ) correspond to the minimum and maximum values of the w values obtained from all actual trajectories respectively. The finally obtained w’ s represents the normalized matching degree of the s - th actual trajectory TR s .

[0113] Step 4.5: Updated cluster centers.

[0114] After matching and classifying the trajectories, for the class where the central trajectory is TD cen , the coordinates of each point of the central trajectory do not change. For the class where the central trajectory is TD’ cen , calculate the average value of the coordinates of each class and select the closest trajectory as the new cluster center to update the corresponding coordinate positions of the original TD’ cen trajectory. Thus, the first iteration is completed.

[0115] Step 4.6: Iterate until the final result is obtained.

[0116] Repeat steps 4.4 and 4.5 until the cluster centers no longer change or the number of clustering times reaches the requirement, that is, output the results after the clustering results reach a stable state.

[0117] Taking Nanjing Lukou Airport as an example, the secondary radar data of the terminal area from July 20 to August 11, 2019 of Nanjing Lukou Airport was used. 14,337 tracks were selected, and after data cleaning and screening, 4,891 arrival tracks and 5,387 departure tracks were obtained. The optimal number of clusters k required for clustering of arrival and departure tracks was obtained as 22 by calculating the sum of squared errors.

[0118] For the takeoff and landing tracks, the silhouette coefficient comparison method was used to compare the advantages and disadvantages of the clustering results of the standard procedure guidance and randomly selected initial cluster centers. It can be seen from the results in Table 1 that the clustering effect obtained by the improved k-means clustering method with the standard flight procedure guidance is better than the original clustering effect.

[0119] Result comparison chart of Table 1

[0120]

[0121] Analyze the clustering results of the arrival and departure tracks in the terminal area of Nanjing Lukou Airport, and the obtained clustering results can be seen Figure 4 . The corresponding cluster centers after clustering of the arrival and departure tracks are as Figure 5 shown.

Claims

1. A method for mining aircraft flight patterns in the airport terminal area based on standard flight procedures, characterized in that It includes the following steps: Step 1: Obtain the takeoff track, landing track, overflight track data and standard flight procedures of departures and arrivals; Step 2: Conduct quality analysis on the obtained data and perform preprocessing; Step 3: Perform coordinate transformation on the track point data and construct a multi-dimensional track feature vector; Step 4: Determine the central track of each type of flight mode and obtain the corresponding flight mode features; The said Step 3 includes the following steps: Step 3.1: Perform coordinate transformation on the track point position information: Convert the longitude data and latitude data returned by the secondary radar into the northeast-up coordinate system, which reflects the distance relationship with the airport and the actual track position information of the aircraft; create a rectangular coordinate system with the airport's navigation station as the coordinate origin, and obtain the e, n, u coordinate values of the northeast-up coordinate system; take the user's coordinate origin (x0, y0, z0) as the coordinate axis origin, and set the position of the coordinate point (x, y, z) in the northeast-up coordinate system as (e, n, u); at the same time, set the longitude-latitude-altitude coordinate point of the coordinate origin as LLA0 = (lon0, lat0, alt0), then the corresponding conversion formula is: Step 3.2: Construct a multi-dimensional track feature vector: The obtained track data includes multi-dimensional features such as longitude, latitude, altitude, speed, and heading. The shape of the track is completely described by constructing the climb rate a and the angle θ; The corresponding calculation methods for constructing the climb rate a and the angle θ are as follows: Among them, v j represents the aircraft speed at this point at time t = j; dist refers to the Euclidean distance calculated between two points; Px j , Py j respectively represent the abscissa and ordinate corresponding to this point; According to the converted coordinates e, n, u, the lifting rate a and the angle θ, as well as the known information such as the speed v and the course γ and other characteristics, the characteristic data P of each trajectory point is obtained j : P j = (e, n, u, a, v, γ, θ) (5) where P j is the trajectory point corresponding to the trajectory at the moment t = j. Formula (5) represents the e, n, u coordinates of the multi-dimensional trajectory point corresponding to the trajectory at the moment t = j, and the seven characteristic parameters of the lift rate, speed, course, and angle. The track TR is a set of track point sequences, j ∈ [1, m] is the track point number, and m is the total number of track points. Then the set of track points changing with time is: TR = {P1, P2,..., P j ,..., P m} (6) Suppose there are a total of h tracks, TD is the set of aircraft tracks, i ∈ [1, h] is the track number, and h is the total number of tracks; the set of aircraft tracks can be represented in the form of the following set: TD = {TR1, TR2,..., TR i ,..., TR h} (7); The said Step 4 includes the following steps: Step 4.1: Search for the track closest to the standard flight procedure from the historical track data; Finding the actual trajectory closest to the standard flight procedure: From all actual trajectories TD B ={TR b1 ,TR b2 ,...,TR bQ}, use the Euclidean distance method to calculate the distances from all standard procedures TD A ={TR a1 ,TR a2 ,...,TR aR}, and find the actual trajectory closest to each standard procedure trajectory from TD B . Denote the set of these trajectories as TD cen ; Step 4.2: Match the actual trajectory with TD cen as follows: Calculate the set TD separately B The dynamic time warping distances between other trajectories in cen and these TD trajectories: δ(P ay ,P bz )=(P ay -P bz ) 2 (8) Among them, P ay is defined as the central trajectory sequence of TR a ={P a1 , P a2 ,..., P ay ,..., P aY}, and P bz is defined as the actual track sequence of TR b ={P b1 , P b2 ,..., P bz ,..., P bZ}. The d(P ay , P bz ) is the DTW distance between two points, and the sum of the DTW distances of track points is used as the relevance of this group of tracks; in the TD cen set, several actual tracks are selected near each track as its matching tracks for the corresponding cluster of TD C ∈TD B , completing the initial clustering of tracks based on standard flight procedures; the actual track TD B is divided into three types of tracks: the cluster track TD cen corresponding to each track in TD cen , the remaining track TD C , and the remaining tracks TD D . A new clustering center track will be selected from the remaining tracks; Step 4.3: Randomly select the remaining initial central tracks; From TD B The remaining trajectories in TD D Randomly select k - r trajectories as the remaining initial center trajectories, denoted as TD' cen , for the other trajectories in TD D to perform matching; the set TD cen and the set TD' cen constitute all the initial center trajectories; the remaining trajectories are assigned to the clusters corresponding to the trajectories in the TD' cen set using the DTW matching principle; Step 4.4: Matching degree analysis: According to the k central trajectories, calculate the DTW distance d(P ay , P bz ) between the actual trajectory and each central trajectory; Store the sub - result calculated by d(P ay , P bz ) in the cumulative distance matrix D(Y, Z), where D(Y, Z) is a matrix with Y rows and Z columns, which is convenient for storing all the results of d(P ay , P bz ). d(Y, Z) is used as the optimal global distance, representing the similarity between two trajectories; Calculate the central trajectory sequence TR of the l-th l Similarity distance from the given actual track sequence, denoted as d l (Y, Z); Introduce w for matching degree analysis: w = min(d l (X, Y)) for each actual trajectory TR s corresponds to the matching degree w of the center trajectory s ; The data is linearly mapped to the interval [0, 1] by using the min-max normalization method: where, min(w s ) and max(w s ) correspond to the minimum and maximum values of the w values obtained from all actual trajectories respectively, and the finally obtained w’ s represents the normalized matching degree of the s-th actual trajectory TR s ; Step 4.5: Updated cluster centers: For the class where the central trajectory is TD cen Each point coordinate of the central trajectory remains unchanged, and the central trajectory is TD’ cen For the class where the central trajectory is TD’, calculate the average value of the coordinates of each class and select the trajectory closest to it as the new cluster center to update the original TD’ cen The corresponding coordinate positions of the trajectory, and thus the first iteration is completed; Step 4.6: Repeat Step 4.4 and Step 4.5 until the clustering center no longer changes or the number of clustering times reaches the requirement, that is, after the clustering result reaches a stable state, output the result.

2. The method for mining aircraft flight patterns in the airport terminal area based on standard flight procedures according to claim 1, wherein The takeoff track, landing track, and overflight track data in Step 1 include information recording time, flight number, aircraft position information, heading, climb rate, and aircraft speed.

3. The method for mining aircraft flight patterns in the airport terminal area based on standard flight procedures according to claim 1, wherein The said Step 2 includes the following steps: Step 2.1: Aircraft track merging: The obtained track data is classified by hour, and the track point information within each hour is a group; Step 2.2: Number the tracks: When the time interval between adjacent track points with the same flight number exceeds 15 minutes, it is defined that these two track points come from different flights; the tracks of each day's flights are numbered and distinguished by the year, month, day plus the track ordinal number. Step 2.3: For duplicate values, missing values, and outliers: For duplicate values, adopt the method of deletion, retain the first set of duplicate data and delete the remaining duplicate data; for missing values, use the method of imputation or deletion for processing; for data with a too small missing rate, use the regression method to fill in the missing data, and for data with a too large missing rate, delete all information of the track point; for outliers, uniformly adopt the method of deletion; for tracks with too few track points and too large interval between two adjacent points, delete them.

Images