An aircraft intention-based flight behavior identification method in an autonomous operation mode

Abstract

The application discloses a flight behavior recognition method based on aircraft intention in an autonomous operation mode. Firstly, aircraft behavior features are extracted from the time and space levels; secondly, an aircraft flight intention model in a typical flight scene is constructed; thirdly, the aircraft behavior features are analyzed by clustering based on a BIRCH clustering model, and the typical behaviors of the aircraft in flight are divided; and finally, the real-time behavior feature values are calculated based on the flight intention model, and online behavior recognition of the aircraft flight is realized. The aircraft behavior features are constructed from the time and space angles, so that the aircraft behavior is more visualized. In addition, the aircraft flight intention model can adapt to dynamic change scenes and is more flexible. Finally, the BIRCH model can adapt to the process of incremental clustering and meet the real-time performance of online behavior recognition.

Description

TECHNICAL FIELD

[0001] The application relates to an aircraft intention-based flight behavior recognition method in an autonomous operation mode and belongs to the behavior recognition field. BACKGROUND

[0002] In the autonomous operation mode, aircraft flight behavior recognition is one of the key technologies for ensuring flight safety and efficiency. In different stages, the flight behaviors of the aircraft are quite different. In the cruising stage, the aircraft usually adopts uniform straight flight, and in the approach stage, there are more turning flights and larger speed changes, and the operation complexity increases dramatically. The autonomous operation increases the degree of freedom of flight, and the risk also increases accordingly. Therefore, flight behavior recognition can provide guidance for the safety control of the aircraft, and behavior prediction can lay a foundation for flight path prediction, thereby providing a safety guide for the safe flight of the autonomous operation aircraft.

[0003] Current aircraft behavior feature recognition usually faces the terminal area flight scene, and there is little flight behavior recognition throughout the entire flight, which is not coherent. In addition, most behavior recognition is based on the calculation of flight path similarity through spatial distortion, thereby obtaining behavior similarity, ignoring the influence of the time characteristics of the aircraft flight on the behavior. Moreover, the mainstream behavior recognition of the aircraft in the civil aviation field is offline calculation, which cannot meet the real-time needs of the autonomous operation mode. Therefore, it is urgent to propose a behavior recognition method of behavior features with comprehensive space-time characteristics, which can adapt to the dynamically changing airspace environment, be applicable to the entire flight, and ensure the timeliness of data updating. SUMMARY

[0004] The application provides an aircraft intention-based flight behavior recognition method in an autonomous operation mode. Through the method, the flight intention of the aircraft can be dynamically updated, the current flight behavior type of the aircraft can be recognized, the prediction accuracy can be improved as a pre-algorithm for flight path prediction, and reference can be provided for flight guidance in the autonomous operation mode. The steps are as follows:

[0005] Step 1: An aircraft flight behavior feature system is constructed from the time and space levels, and the ADS-B data of the airport OD pair is processed.

[0006] Step 2: An aircraft variable flight intention model of the aircraft in a conventional scene and an abnormal scene is constructed, and a next key intention point is obtained.

[0007] Step 3: A BIRCH clustering algorithm is adopted to cluster the flight behavior based on the space-time characteristics of the aircraft, and typical flight behaviors of the aircraft are divided.

[0008] Step 4: The online flight behavior feature value of the aircraft is calculated based on the aircraft flight intention model, the BIRCH clustering algorithm is adaptively adjusted, and real-time recognition of the aircraft behavior is realized.

[0009] Step 1 specifically includes:

[0010] Step 1.1: Build an aircraft flight behavior feature system from time and space dimensions, select the turn angle, current speed, and deviation distance as the aircraft flight space-time features, where the turn angle and deviation distance are defined as follows:

[0011] ① Turn angle: the spatial angle formed by the aircraft flying through two consecutive segments.

[0012]

[0013] In the formula: - the angle between consecutive segments;

[0014] - the position vector of the previous segment;

[0015] - the position vector of the next segment.

[0016] ② Deviation distance: the distance of the aircraft cutting into the planned route.

[0017]

[0018] In the formula: d o is the projection distance of the aircraft cutting into the segment, a is the angle between the aircraft heading and the cutting segment, p o represents the aircraft, p M represents the intersection of the aircraft heading extension line and the cutting segment;

[0019] Step 1.2: Obtain flight plan information, and process ADS-B timestamps of airport OD pairs at equal intervals and smooth the track point features using Kalman filtering.

[0020] Step 1.3: To eliminate the influence of different features between dimensions, use Z-score standardization features to speed up the convergence speed, the formula is as follows:

[0021]

[0022] Where, x norm is the feature standardization value, x int is the feature initial value, x is the feature mean, and σ is the feature standard deviation x int .

[0023] Step 2 specifically includes:

[0024] Step 2.1: Use a binary tree search algorithm to obtain the nearest waypoint W2 from the current aircraft and its previous and next two waypoints W1, W3.

[0025] Step 2.2: Establish an aircraft flight intention model under a normal flight scenario, and obtain the next key point of the aircraft through the angle θ. First, determine the inside and outside of the flight segment WP1-WP2-WP3, and calculate the cross product of two vectors to determine the inside and outside to which the current aircraft belongs. The calculation is as follows:

[0026]

[0027] Step 2.3: Determine the membership flight segment and intention point. When the aircraft is outside the flight segment, the next key point is calculated as follows:

[0028]

[0029] When the aircraft is inside the flight segment, the next key point is calculated as follows:

[0030]

[0031] Where L1 is the vector formed by the aircraft and the nearest waypoint, L2 is the vector formed by the aircraft and the next waypoint of the nearest waypoint, and τ1 and τ2 are the constraint angles for determining the next key point.

[0032] Step 2.4: Establish an aircraft flight intention model under an abnormal flight scenario, and update the next key point to the trajectory planning point according to the new flight plan established by the abnormal event triggered trajectory planning.

[0033] Step 3 specifically includes:

[0034] Step 3.1: Obtain n m-dimensional ADS-B flight data samples X = {X1, X2, …, Xn}, and construct a clustering feature tree. n}, construct a clustering feature tree.

[0035] Step 3.2: On the clustering feature tree, construct a clustering feature CF triple (N, LS, SS)

[0036]

[0037] In the formula: X i is the feature vector of the i-th sample in the CF, x im is the m-th feature of the i-th sample, and N represents the sample size.

[0038] Step 3.3: There is no node on the initial clustering feature tree, and the aircraft samples are read in turn and added to the CF triple.

[0039] Step 3.4: When the maximum capacity limit of the leaf node is exceeded or the branch factor is exceeded, the combination and splitting operations of the execution node will be performed.

[0040] Step 3.5: After all sample readings are completed, the cluster feature tree completes preliminary clustering. Then, the K-means algorithm is used to further cluster the triplets in these preliminary clusters.

[0041] Step 3.6: The silhouette coefficient is used to determine the optimal number of clusters, which is calculated as follows:

[0042]

[0043] In the formula: a(i) is the average distance of sample i and other sample points in the cluster. b(i) is the average distance of sample i to each sample point in the nearest cluster.

[0044] Step 4 specifically includes:

[0045] Step 4.1: Real-time addition of new flight data points, based on the aircraft flight intent model to obtain the next key point, taking the point as the last point of the turn angle, calculating the online flight behavior feature value of the aircraft.

[0046] Step 4.2: Input the calculated feature value into the cluster feature tree for dynamic adjustment, adaptive BIRCH clustering algorithm, further training the model, realizing incremental clustering, realizing real-time recognition of aircraft behavior.

[0047] The beneficial effects achieved by the present application are: establishing aircraft flight behavior features from two aspects of time and space, more comprehensively summarizing the time and space information of aircraft flight; establishing aircraft flight behavior variable intent models under different scenarios, which can adapt to the dynamically changing airspace environment and ensure the effectiveness of recognition; using the BIRCH clustering algorithm to comprehensively divide the aircraft behavior features, obtaining typical flight behaviors, combining flight intent to calculate online features, and realizing adaptive incremental clustering mode, realizing dynamic behavior recognition update. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The flowchart of the method of the present application is shown in Figure 1.

[0049] Figure 2 The turn angle calculation schematic diagram is shown in Figure 2.

[0050] Figure 3 The yaw distance calculation schematic diagram is shown in Figure 3.

[0051] Figure 4 The aircraft leg matching search schematic diagram is shown in Figure 4.

[0052] Figure 5Fig. 1 is a lateral judgment diagram for an aircraft within a flight segment, wherein the upper left is an aircraft outside the flight segment and not flying over the nearest waypoint, the upper right is an aircraft outside the flight segment and having flown over the nearest waypoint, the lower left is an aircraft inside the flight segment and not flying over the nearest waypoint, and the lower right is an aircraft inside the flight segment and having flown over the nearest waypoint;

[0053] Figure 6 Fig. 6 is a key point capture visualization diagram for an example intention;

[0054] Figure 7 Fig. 7 is an intention change schematic diagram in an abnormal flight scenario

[0055] Figure 8 Fig. 8 is a BIRCH clustering feature tree schematic diagram;

[0056] Figure 9 Fig. 9 is a BIRCH leaf node splitting and combination schematic diagram;

[0057] Figure 10 Fig. 10 is a contour coefficient calculation result;

[0058] Figure 11 Fig. 11 is a BIRCH-based typical flight behavior division, wherein (a) is right turn deceleration, (b) is constant speed straight flight, (c) is left turn constant speed, (d) is deceleration straight flight, (e) is right turn acceleration, (f) is right turn constant speed, (g) is acceleration straight flight, (h) is left turn acceleration, and (i) is left turn deceleration; Embodiments

[0059] To make the purpose, technical solutions and points of the present application more clear, OD pair ADS-B data from Chengdu Shuangliu Airport ZUUU to Hangzhou Xiaoshan Airport ZSHC is selected to perform flight behavior recognition. The present application is further described below in combination with the drawings.

[0060] The present application proposes a flight behavior recognition method based on aircraft intention in an autonomous operation mode, as shown in Fig. 1: the specific implementation steps are as follows: Figure 1

[0061] Step 1: constructing an aircraft flight behavior feature system from the time and space layers, and processing the ADS-B data of the OD pair of the airport.

[0062] Step 1.1: selecting the turn angle, current speed and yaw distance as the aircraft flight space-time features, wherein the turn angle and yaw distance are calculated as follows:

[0063] ① Turn angle: the spatial angle formed by the aircraft flying over two consecutive flight segments, as shown in Fig. 2. Figure 2 .

[0064]

[0065] ​In the formula: — the included angle of consecutive segments;

[0066] — the position vector of the previous segment;

[0067] — the position vector of the next segment.

[0068] ② The yaw distance: the distance of the aircraft cutting into the planned route as soon as possible, such as Figure 3 .

[0069]

[0070] In the formula: d o is the projection distance of the aircraft cutting into the segment, α is the included angle between the aircraft heading and the cutting-in segment, p o represents the aircraft, p M represents the intersection of the aircraft heading extension line and the cutting-in segment;

[0071] Step 1.2: Obtain the flight plan information from ZUUU to ZSHC, and process the ADS-B time stamps of the OD pair of the two airports at 15s intervals. The longitude, latitude, height, speed and heading features of the smoothed track points are smoothed by Kalman filtering to improve the accuracy of behavior recognition.

[0072] Step 1.3: In order to eliminate the influence of different dimensions between different features, Z-score standardization features are used to speed up the convergence speed, and the formula is as follows:

[0073]

[0074] Where, x norm is the feature standardization value, x int is the feature initial value, is the feature mean, and σ is the feature standard deviation x int .

[0075] Step 2: Construct the aircraft variable flight intention model of the aircraft under the normal scene and the abnormal scene, and obtain the next key intention point.

[0076] Step 2.1: Use the binary tree search algorithm to obtain the nearest waypoint W2 and its previous and next two waypoints W1, W3 from the current aircraft, as shown in Figure 4 The blue area range, the aircraft shows the segment WP1-WP2-WP3, and in this range, the nearest point of the aircraft is always WP2.

[0077] Step 2.2: Establish aircraft flight intention model under normal flight scene, get next key point of aircraft through angle θ. Determine the inside and outside of the aircraft by calculating the cross product of two groups of vectors, as follows:

[0078]

[0079] Step 2.3: Determine the aircraft's current membership in the segment and the intention point, as follows: Figure 5 When the aircraft is outside the segment, select the division angle as 90°, and the next key point is calculated as follows:

[0080]

[0081] When the aircraft is inside the segment, select the division angle as 45°, and the next key point is calculated as follows:

[0082]

[0083] Where L1 is the vector formed by the aircraft and the nearest waypoint, and L2 is the vector formed by the aircraft and the next waypoint of the nearest waypoint.

[0084] The implementation is shown in the following: Figure 6

[0085] Step 2.4: When entering an abnormal flight scene, update the next key point to the trajectory planning point according to the new flight plan established by the abnormal event triggered trajectory planning, as follows: Figure 7

[0086] Step 3: Use BIRCH clustering algorithm to cluster the flight behavior based on the spatio-temporal characteristics of the aircraft, and divide the typical flight behavior of the aircraft.

[0087] Step 3.1: There are 60286 3-dimensional behavior feature data samples X = {X1, X2, X3}, and a clustering feature tree CFTree is constructed, as follows: Figure 8

[0088] Step 3.2: On the CF Tree, construct a clustering feature CF triple (N, LS, SS)

[0089]

[0090] Where: X i represents the 3-dimensional feature vector of each flight feature sample in CF, x im represents the mth feature of the ith sample, and N represents the number of samples in CF.

[0091] Step 3.3: There are no nodes on the initial CF Tree, and the feature samples are read in turn and added to the CF triple.​​​

[0092] Step 3.4: When the maximum CL number L of the leaf node is exceeded or the branch factor B is exceeded, the node group splits, as shown in Figure 9 .

[0093] Step 3.5: After reading the sample, the CF Tree completes the clustering of the sample once, forms several CFs, and then uses Kmeans to perform secondary clustering on the CF triplets of the leaf nodes.

[0094] Step 3.6: The contour coefficient calculation result is as shown in Figure 10 , and the optimal clustering number is determined to be 9, and 9 typical flight behaviors are obtained, as shown in Figure 11 .

[0095] Step 4: Based on the aircraft flight intention model, the online flight behavior characteristic value of the aircraft is calculated, the adaptive BIRCH clustering algorithm is used, and the real-time recognition of the aircraft behavior is realized.

[0096] Step 4.1: Real-time addition of new flight feature data points, based on the aircraft flight intention model to obtain the next key point, taking the point as the last point of the turn angle, and calculating the online flight behavior characteristic value of the aircraft.

[0097] Step 4.2: Through dynamic adjustment of the clustering feature tree, the adaptive BIRCH clustering algorithm is used to further train the model, realize incremental clustering, and realize real-time recognition of the aircraft behavior, as shown in Table 1.

[0098] Table 1

[0099]

[0100] The above is only an embodiment of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application is included in the scope of the claims of the application to be approved.

Claims

1. A method for recognizing flight behavior based on aircraft intent in autonomous operation mode, characterized in that: Includes the following steps: Step 1: Construct an aircraft flight behavior characteristic system from the time and space levels, and process the ADS-B data of airport OD pairs; Step 2: Construct a variable flight intent model for the aircraft in normal and abnormal scenarios to obtain the next key intent point; Step 3: Use the BIRCH clustering algorithm to cluster flight behaviors based on the spatiotemporal characteristics of aircraft and classify typical flight behaviors of aircraft. Step 4: Calculate the online flight behavior feature values ​​of the aircraft based on the aircraft variable flight intention model, and use the adaptive BIRCH clustering algorithm to realize real-time recognition of aircraft behavior; Step 2 specifically includes: Step 2.1: Use a binary tree search algorithm to obtain the waypoint WP2 that is closest to the current aircraft and the two waypoints WP1 and WP3 before and after it; Step 2.2: Establish an aircraft flight intent model under normal flight scenarios, and obtain the next key point of the aircraft through angle θ; first, determine the inner and outer sides of the flight segment WP1-WP2-WP3, and calculate the cross product of the two sets of vectors to determine the inner and outer sides to which the current aircraft belongs, as follows: Step 2.3: Determine the flight segment and the intended point; when the aircraft is outside the flight segment, the next key point is calculated as follows: When the aircraft is inside the flight segment, the next critical point is calculated as follows: Where L1 is the vector formed by the aircraft and the nearest waypoint, L2 is the vector formed by the aircraft and the next waypoint of the nearest waypoint, and τ1 and τ2 are the constraint angles for determining the next key point; Step 2.4: Establish an aircraft flight intent model under abnormal flight scenarios, and update the next key point as the trajectory planning point based on the new flight plan established by the trajectory planning triggered by the anomaly.

2. The method for flight behavior recognition based on aircraft intent in autonomous operation mode according to claim 1, characterized in that: Step 1 specifically includes: Step 1.1: Construct an aircraft flight behavior characteristic system from the temporal and spatial levels, selecting turning angle, current speed, and yaw distance as the aircraft's spatiotemporal characteristics. The turning angle and yaw distance are defined as follows: ①Turning angle: The spatial angle formed when an aircraft flies through two consecutive segments; In the formula: —Angle between consecutive flight segments; —Previous segment position vector; —Position vector for the next flight segment; ②Yaw distance: The distance at which an aircraft can most quickly enter its planned route; In the formula: d o Let p be the projected distance of the aircraft's entry segment, α be the angle between the aircraft's heading and the entry segment, and p be the projected distance of the aircraft's entry segment. o Indicates aircraft, p M This indicates the intersection of the aircraft's extended heading and the entry segment; Step 1.2: Obtain flight plan information, perform equal-interval processing on the ADS-B timestamps of airport OD pairs, and use Kalman filtering to smooth the track point features; Step 1.3: To eliminate the influence of different features' dimensions, Z-score normalization is used to accelerate convergence. The formula is as follows: Where, x norm For the standardized value of the feature, x int As the initial value of the feature, Let σ be the characteristic mean and σ be the characteristic standard deviation. int .

3. The method for flight behavior recognition based on aircraft intent in autonomous operation mode according to claim 2, characterized in that: Step 3 specifically includes: Step 3.1: Obtain n m-dimensional ADS-B flight data samples X = {X1, X2, ..., X...} n }, construct its clustering feature tree; Step 3.2: Construct cluster feature CF triples (N, LS, SS) on the cluster feature tree. In the formula: X i Let x be the feature vector of the i-th sample in CF. im Let m be the m-th feature of the i-th sample, and N represent the sample size; Step 3.3: The initial cluster feature tree has no nodes. Read the aircraft samples in sequence and add them to the CF triples. Step 3.4: When the maximum capacity limit of a leaf node or the branching factor is exceeded, node combination and splitting operations will be performed; Step 3.5: After all samples have been read, the clustering feature tree completes the initial clustering, and then the K-means algorithm is used to perform further secondary clustering on the triples in these initial clusters; Step 3.6: Determine the optimal number of clusters using the silhouette coefficient, calculated as follows: In the formula: a(i) is the mean distance between sample i and other sample points in the cluster, and b(i) is the mean distance between sample i and each sample point in the nearest cluster.

4. The method for flight behavior recognition based on aircraft intent in autonomous operation mode according to claim 1, characterized in that: Step 4 specifically includes: Step 4.1: Add new flight data points in real time, obtain the next key point based on the aircraft flight intent model, use this point as the last point of the turning angle, and calculate the online flight behavior feature value of the aircraft; Step 4.2: Input the calculated feature values ​​into the clustering feature tree for dynamic adjustment, adaptive BIRCH clustering algorithm, further train the model, realize incremental clustering, and realize real-time recognition of aircraft behavior.

Images