A data-driven terminal area traffic complexity assessment and prediction method

Abstract

The application discloses a data-driven terminal area traffic complexity evaluation and prediction method, which comprises the following steps: obtaining preprocessed track data; obtaining track data of each space-time grid; calculating macroscopic and microscopic evaluation index data of traffic complexity; dividing traffic complexity levels of each space-time grid in different time periods; and constructing a terminal area traffic complexity level prediction model based on CNN to predict future traffic complexity levels. The application has the advantages that: in the complexity index construction aspect, the macroscopic and microscopic features of air traffic complexity are considered simultaneously, and a clustering method is used to divide the complexity levels, so that the results have strong objectivity and universality; through space-time grid division of the terminal area, fine evaluation of the terminal area traffic complexity is realized; and the traffic complexity prediction model simultaneously considers the time correlation of traffic complexity evolution and the spatial correlation between adjacent grids, so that the prediction accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of air traffic management technology, and specifically relates to a data-driven method for assessing and predicting terminal area traffic complexity. Background Technology

[0002] With the continuous growth of air traffic volume, terminal area traffic has become increasingly complex, leading to more severe flight conflict situations and a surge in the workload of air traffic controllers. Highly complex airspace often exhibits significant traffic disorder. On the one hand, this can easily trigger flight conflicts between aircraft, endangering the lives of passengers and crew and causing substantial economic losses. On the other hand, to avoid flight conflicts, aircraft may employ maneuvers such as detours, circling, and deceleration, thereby increasing the time aircraft spend in the terminal area and reducing traffic efficiency. Therefore, scientifically assessing and accurately predicting the complexity of terminal area traffic is crucial for ensuring air traffic safety and improving terminal area traffic operation efficiency.

[0003] Methods for assessing complex air traffic situations typically construct complexity indicators based on the relative positions of individual aircraft, such as relative distance and relative speed, and then have controllers assign a complexity level. This method is highly susceptible to the influence of individual controllers' experience, mood, and fatigue; different controllers or even the same controller under different conditions may give different results. Therefore, this method has high uncertainty and poor objectivity and universality.

[0004] Methods for predicting complex air traffic situations generally fall into two categories: prediction using modeling and simulation, and prediction using data-driven methods. Air traffic systems are composed of many different entities, and these elements interact with each other, forming a nonlinear, aggregated dynamic characteristic, which poses a challenge to accurate delay prediction. Modeling and simulation-based prediction methods primarily establish mathematical models, requiring sophisticated use of many model parameters and incorporating unrealistic assumptions. Data-driven methods often directly utilize historical complexity index data to predict the overall complexity of a terminal area, failing to provide refined predictions of complexity across different regions within the terminal area. Furthermore, they mainly consider the temporal correlation of complexity, rarely considering the correlation between different regions within the terminal area.

[0005] In summary, the current research has the following main problems: (1) The current air traffic complexity assessment mainly considers the relative positional relationship between individual aircraft, and the complexity level is given by the controller, which lacks objectivity and universality; (2) Existing research has not achieved refined prediction of the complexity of different areas within the terminal area; (3) Existing complexity prediction research lacks consideration of the temporal and spatial correlation characteristics of traffic complexity. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to provide a data-driven method for assessing and predicting traffic complexity in terminal areas.

[0007] To achieve the above objectives, the data-driven terminal area traffic complexity assessment and prediction method provided by this invention includes the following steps performed in sequence:

[0008] 1) Collect the actual arrival and departure flight trajectory data of the terminal area to be tested; then convert the trajectory data from the latitude and longitude coordinate system to the rectangular coordinate system; finally clean the abnormal data in the trajectory data and fill in the missing data, thereby preprocessing the trajectory data and obtaining the preprocessed trajectory data.

[0009] 2) Divide the terminal area airspace into a four-dimensional spatiotemporal grid, and then use the preprocessed track data obtained in step 1) to allocate each track point data to the corresponding four-dimensional spatiotemporal grid according to its spatial location and time period, so as to obtain the track data of each spatiotemporal grid.

[0010] 3) Using the trajectory data of each spatiotemporal grid obtained in step 2), calculate the macro and micro assessment index data of traffic complexity;

[0011] 4) Using the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), the K-Means clustering algorithm is used to classify the traffic complexity level of each spatiotemporal grid at different time periods;

[0012] 5) Construct a terminal area traffic complexity level prediction model based on CNN. The input of this model is the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), and the output is the traffic complexity level of each spatiotemporal grid obtained in step 4). By learning the spatiotemporal evolution law of air traffic complexity through the terminal area traffic complexity level prediction model, the future traffic complexity level can be predicted based on historical traffic complexity index data.

[0013] In step 1), the actual arrival and departure flight track data of the terminal area to be tested are collected; then the track data is converted from the latitude and longitude coordinate system to the rectangular coordinate system; finally, abnormal data in the track data is cleaned and missing data is filled in. The specific method for preprocessing the track data and obtaining the preprocessed track data is as follows:

[0014] 1.1) Track data acquisition

[0015] Collect the latitude and longitude coordinates, altitude, speed, acceleration, and heading data of each track point in the actual arrival and departure flight track data of the terminal area to be tested;

[0016] 1.2) Coordinate Transformation

[0017] Convert the latitude and longitude coordinates of all track points collected in step 1.1) into Cartesian coordinates, using the following formula:

[0018]

[0019]

[0020] Where a is the semi-major axis of the reference ellipsoid, b is the semi-minor axis of the reference ellipsoid, e is the first eccentricity, e' is the second eccentricity, and the latitude and longitude λ0 of the center point of the zero meridian are used. As the origin of the projection, x0, y0 are the coordinates of the midpoint of the selected range. These are the longitude and latitude coordinates of the track point, respectively, (X N ,Y N (x, y) are the Mercator projection coordinates of the transformed track point, and (x, y) are the Cartesian coordinates of the transformed track point.

[0021] 1.3) Deletion of abnormal waypoints

[0022] Calculate the angle between any three consecutive track points on each track after the coordinate transformation obtained in step 1.2). If the angle is acute, it indicates that the middle track point among the three track points is an abnormal track point and it is deleted.

[0023] 1.4) Filling in missing waypoints

[0024] Linear interpolation is used to interpolate each track in step 1.3) at 1-second intervals to fill in missing track points; finally, the preprocessed track data is obtained.

[0025] In step 2), the terminal area airspace is divided into a four-dimensional spatiotemporal grid. Then, using the preprocessed track data obtained in step 1), each track point data is assigned to the corresponding four-dimensional spatiotemporal grid according to its spatial location and time period. The specific method for obtaining the track data of each spatiotemporal grid is as follows:

[0026] 2.1) Spatial grid division

[0027] The terminal area airspace is divided into m and n equal parts along the x and y axes in a Cartesian coordinate system, respectively, thus dividing the horizontal projection surface of the terminal area airspace into m×n adjacent horizontal grids. Then, in the vertical direction, it is divided into r equal parts by unit length, thus dividing the terminal area airspace into m×n×r three-dimensional spatial grids, each three-dimensional spatial grid representing a specific spatial region. Then, each track point in the preprocessed track obtained in step 1) is assigned to its respective spatial grid according to its spatial position.

[0028] 2.2) Time grid partitioning

[0029] Using Δt as the time interval, the track point data of each spatial grid obtained in step 2.2) are allocated to the corresponding time interval according to their time, thereby dividing the terminal area airspace into a four-dimensional spatiotemporal grid and obtaining the track data of each spatiotemporal grid.

[0030] In step 3), the specific method for calculating the macro and micro assessment indicators of traffic complexity using the trajectory data of each spatiotemporal grid obtained in step 2) is as follows:

[0031] 3.1) Calculation of Macro-level Assessment Indicators for Traffic Complexity

[0032] Using the track data of each spatiotemporal grid obtained in step 2.2), calculate the average velocity of all track points within any spatiotemporal grid, and use this average velocity as the average velocity of that spatiotemporal grid, denoted as v. ins Calculate the average acceleration of all track points within any spatiotemporal grid, and use this as the average acceleration of that spatiotemporal grid, denoted as acc. mean Calculate the standard deviation of the velocities of all track points within any spatiotemporal grid, and use this as the velocity standard deviation of that spatiotemporal grid, denoted as std. v Calculate the number of aircraft passing through any spatiotemporal grid, which is taken as the traffic flow of that spatiotemporal grid, denoted as q. grid ;

[0033] 3.2) Calculation of micro-level assessment indicators for traffic complexity

[0034] The specific method is as follows:

[0035] In a plane, the two-dimensional coordinates of the two aircraft are (x, y, y) and (x, y, y) respectively. i ,y i ) and (x j ,y j Their speeds are respectively and The vector connecting the spatial positions of the two aircraft is Define pi,j The determination coefficient for two aircraft flying on the same route is calculated using the following formula:

[0036]

[0037] Based on the spatial coordinates and velocity vectors of the two aircraft, the following two-dimensional linear equation can be established:

[0038]

[0039] Where (x) i ,y i ),(x j ,y j ) represent the spatial coordinates of aircraft i and aircraft j at a certain moment, respectively, and k i ,k j d represents the ratio of the y-axis and x-axis components of the velocity vectors of the two aircraft, respectively. i ,d j These represent the x-intercepts of aircraft i and aircraft j, respectively.

[0040] If a solution exists by combining the equations of the two lines above, then the intersection point f can be determined. i,j =(x f ,y f The line vectors connecting the intersection points of the coordinates of the two aircraft and the straight line are respectively... Define q i,j The cross-determination coefficient between the two aircraft is calculated using the following formula:

[0041]

[0042] If p i,j =1, which means that the two aircraft are flying along the same route; if q i,j =1 indicates that the two aircraft are flying along different routes and intersecting, at which point the speed at which they intersect is c. i,j The calculation formula is as follows:

[0043]

[0044] Potential conflict time t between the two aircraft i,j The calculation formula is as follows:

[0045]

[0046] When c i,j >0, calculate the coordinates (x) of the intersection conflict point. i,j ,y i,j The formula is as follows:

[0047]

[0048] Then the conflict risk value of a certain spatiotemporal grid in the c-th time interval. The calculation formula is as follows:

[0049]

[0050] Among them, t i,j,c This represents the potential conflict time for aircraft i,j in the c-th time interval;

[0051] Furthermore, the total conflict risk value F of this spatiotemporal grid over M time intervals is... CR for:

[0052]

[0053] In step 4), the specific method for using the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3) to classify the traffic complexity level of each spatiotemporal grid for different time periods using the K-Means clustering algorithm is as follows:

[0054] 4.1) Standardize the traffic complexity assessment index data obtained in step 3), using the following formula:

[0055]

[0056] Where: Q is the traffic complexity assessment index data to be standardized, μ is the mean of the traffic complexity assessment index, and σ is the standard deviation of the traffic complexity assessment index.

[0057] 4.2) Using the traffic complexity assessment index of each time period of the standardized spatiotemporal grid as a sample, the traffic complexity level of each time period of the spatiotemporal grid is divided into four levels: normal, slightly complex, moderately complex and severely complex, and labeled with the numbers 1, 2, 3 and 4 respectively.

[0058] In step 5), the terminal area traffic complexity level prediction model is constructed based on CNN. The input of this model is the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), and the output is the traffic complexity level of each spatiotemporal grid obtained in step 4). The specific method for predicting the future traffic complexity level based on historical traffic complexity index data by learning the spatiotemporal evolution law of air traffic complexity through the terminal area traffic complexity level prediction model is as follows:

[0059] 5.1) The traffic complexity assessment index data of each spatiotemporal grid obtained in step 4.1) and the traffic complexity level of the spatiotemporal grid obtained in step 4.2) constitute a sample set; each spatiotemporal grid corresponds to one sample, the feature data of which is the traffic complexity assessment index data of the spatiotemporal grid, and the label data is the traffic complexity level of the spatiotemporal grid; then 80% of the sample set is used as the training dataset, and the remaining 20% ​​is used as the test dataset.

[0060] 5.2) Input the samples from the above training dataset into the CNN for training. The CNN architecture is CNN1→CNN2→ReLU→Pooling→Softmax, where CNN1 and CNN2 are training layers used to select features, learn and train on the samples, and fit the traffic complexity assessment index data input in the previous step. CNN1 and CNN2 use three-dimensional convolution kernels, and the convolution operation function is Y=f(X×K+b), where X is the input K-dimensional convolution kernel, b is the bias, and f is the ReLU activation function. The first convolution layer is K1∈R 3×3×3×32 Output Y1∈R m×n×r×32 The second convolutional layer is K2∈R 5×5×5×64 Output Y2∈R m ×n×r×64 Then, a max pooling operation is performed, with a pooling kernel S = 2 × 2 × 2 and a stride of 2. The first pooling layer is Y1 → Y′1, and the second pooling layer is Y2 → Y′2. Finally, the training results are classified after passing through a Softmax layer.

[0061] 5.3) Backpropagation is performed on the above CNN using the mean squared error loss function. During this process, relevant parameters in the neurons are updated. The cross-entropy loss function is used to measure the difference between the predicted probability and the true label, and the formula is as follows: Use the Adam optimizer to update network parameters to minimize the cross-entropy loss function. The training iteration termination condition is determined by calculating whether the loss result reaches the preset value.

[0062] 5.4) When the iteration termination condition is met, the CNN training is completed, thus obtaining the terminal area traffic complexity level prediction model. Then, the samples in the above test dataset are input into the prediction model to obtain the traffic complexity level prediction value.

[0063] 5.5) Compare the predicted traffic complexity level obtained in the previous step with the actual traffic complexity level to calculate the prediction accuracy, as shown in the following formula:

[0064]

[0065] Where N is the total number of samples in the sample set, Y i p Y is the predicted traffic complexity level for the i-th sample. i a Let be the actual value of the traffic complexity level for the i-th sample.

[0066] The data-driven terminal area traffic complexity assessment and prediction method provided by this invention has the following beneficial effects:

[0067] The terminal area traffic complexity assessment and prediction method constructed in this invention considers both macroscopic and microscopic characteristics of air traffic complexity in the construction of complexity indicators, and uses clustering methods to classify complexity levels, resulting in strong objectivity and universality. By dividing the terminal area into spatiotemporal grids, a refined assessment of terminal area traffic complexity is achieved. The traffic complexity prediction model considers both the temporal correlation of traffic complexity evolution and the spatial correlation between adjacent grids, improving prediction accuracy.

[0068] The data used in this invention is easy to collect, and through big data mining methods, the evolutionary patterns of complexity can be obtained quickly and conveniently, and the future complex situation can be predicted, which has practical operational value.

[0069] The terminal area traffic complexity assessment and prediction method constructed in this invention provides a new modeling approach for the research field of air traffic complexity in China, enriches the research methods in the field of complexity assessment and prediction, and the complexity prediction results can provide decision-making basis for airports, airlines, and air traffic control. Attached Figure Description

[0070] Figure 1 This is a flowchart of the data-driven terminal area traffic complexity assessment and prediction method provided by the present invention.

[0071] Figure 2 This is a diagram showing the effect of filling in missing track points in an embodiment of the present invention.

[0072] Figure 3 This is a time series diagram of various traffic complexity assessment indicators in this embodiment of the invention.

[0073] Figure 4 This is a schematic diagram of the traffic complexity level classification results for different time periods in each spatiotemporal grid in an embodiment of the present invention.

[0074] Figure 5 (a) and (b) are schematic diagrams of the predicted and actual values ​​of traffic complexity level in the embodiments of the present invention, respectively. Detailed Implementation

[0075] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0076] like Figure 1 As shown, the data-driven terminal area traffic complexity assessment and prediction method provided by this invention includes the following steps performed in sequence:

[0077] 1) Collect the actual arrival and departure flight trajectory data of the terminal area to be tested; then convert the trajectory data from the latitude and longitude coordinate system to the rectangular coordinate system; finally clean the abnormal data in the trajectory data and fill in the missing data, thereby preprocessing the trajectory data and obtaining the preprocessed trajectory data.

[0078] The specific method is as follows:

[0079] 1.1) Track data acquisition

[0080] Collect the latitude and longitude coordinates, altitude, speed, acceleration, and heading data of each track point in the actual arrival and departure flight track data of the terminal area to be tested;

[0081] 1.2) Coordinate Transformation

[0082] Convert the latitude and longitude coordinates of all track points collected in step 1.1) into Cartesian coordinates, using the following formula:

[0083]

[0084]

[0085] Where a is the semi-major axis of the reference ellipsoid, b is the semi-minor axis of the reference ellipsoid, e is the first eccentricity, e' is the second eccentricity, and the latitude and longitude λ0 of the center point of the zero meridian are used. As the origin of the projection, x0, y0 are the coordinates of the midpoint of the selected range. These are the longitude and latitude coordinates of the track point, respectively, (X N ,Y N (x, y) are the Mercator projection coordinates of the transformed track point, and (x, y) are the Cartesian coordinates of the transformed track point.

[0086] 1.3) Deletion of abnormal waypoints

[0087] Calculate the angle between any three consecutive track points on each track after the coordinate transformation obtained in step 1.2). If the angle is acute, it indicates that the middle track point among the three track points is an abnormal track point and it is deleted.

[0088] 1.4) Filling in missing waypoints

[0089] Linear interpolation is used to interpolate each track in step 1.3) at 1-second time intervals to fill in missing track points. Let the coordinates of two consecutive track points on a track be (x1, y1, z1, t1) and (x2, y2, z2, t2), and t2 - t1 > 1s. The interpolated track point coordinates are (x′, y′, z′, t′), where t1 < t′ < t2. Then:

[0090]

[0091] Finally, the preprocessed track data is obtained.

[0092] 2) Divide the terminal area airspace into a four-dimensional spatiotemporal grid, and then use the preprocessed track data obtained in step 1) to allocate each track point data to the corresponding four-dimensional spatiotemporal grid according to its spatial location and time period, so as to obtain the track data of each spatiotemporal grid.

[0093] The specific method is as follows:

[0094] 2.1) Spatial grid division

[0095] The terminal area airspace is divided into m and n equal parts along the x and y axes in a Cartesian coordinate system, respectively, thus dividing the horizontal projection surface of the terminal area airspace into m×n adjacent horizontal grids. Then, in the vertical direction, it is divided into r equal parts by unit length, thus dividing the terminal area airspace into m×n×r three-dimensional spatial grids, each three-dimensional spatial grid representing a specific spatial region. Then, each track point in the preprocessed track obtained in step 1) is assigned to its respective spatial grid according to its spatial position.

[0096] 2.2) Time grid partitioning

[0097] Using Δt as the time interval, the track point data of each spatial grid obtained in step 2.2) are allocated to the corresponding time interval according to their time, thereby dividing the terminal area airspace into a four-dimensional spatiotemporal grid and obtaining the track data of each spatiotemporal grid.

[0098] 3) Using the trajectory data of each spatiotemporal grid obtained in step 2), calculate the macro and micro assessment index data of traffic complexity;

[0099] The specific method is as follows:

[0100] 3.1) Calculation of Macro-level Assessment Indicators for Traffic Complexity

[0101] Using the track data of each spatiotemporal grid obtained in step 2.2), calculate the average velocity of all track points within any spatiotemporal grid, and use this average velocity as the average velocity of that spatiotemporal grid, denoted as v. insCalculate the average acceleration of all track points within any spatiotemporal grid, and use this as the average acceleration of that spatiotemporal grid, denoted as acc. mean Calculate the standard deviation of the velocities of all track points within any spatiotemporal grid, and use this as the velocity standard deviation of that spatiotemporal grid, denoted as std. v Calculate the number of aircraft passing through any spatiotemporal grid, which is taken as the traffic flow of that spatiotemporal grid, denoted as q. grid ;

[0102] 3.2) Calculation of micro-level assessment indicators for traffic complexity

[0103] Using the trajectory data of each spatiotemporal grid obtained in step 2.2), based on the spatial position and velocity vector data of the aircraft, and assuming uniform linear motion, the potential collision points between any two aircraft are predicted, and the potential collision time is calculated. The reciprocal of the potential collision time is used to characterize the collision risk. If there is no tendency for the two aircraft to intersect, their collision risk value is zero. The collision risk values ​​between any two aircraft within the spatiotemporal grid are accumulated to obtain the collision risk value of that spatiotemporal grid, denoted as F. CR .

[0104] The specific method is as follows:

[0105] In a plane, the two-dimensional coordinates of the two aircraft are (x, y, y) and (x, y, y) respectively. i ,y i ) and (x j ,y j Their speeds are respectively and The vector connecting the spatial positions of the two aircraft is Define p i,j The determination coefficient for two aircraft flying on the same route is calculated using the following formula:

[0106]

[0107] Based on the spatial coordinates and velocity vectors of the two aircraft, the following two-dimensional linear equation can be established:

[0108]

[0109] Where (x) i ,y i ),(x j ,y j ) represent the spatial coordinates of aircraft i and aircraft j at a certain moment, respectively, and k i ,k j d represents the ratio of the y-axis and x-axis components of the velocity vectors of the two aircraft, respectively. i ,d j These represent the x-intercepts of aircraft i and aircraft j, respectively.

[0110] If a solution exists by combining the equations of the two lines above, then the intersection point f can be determined. i,j =(x f ,y f The line vectors connecting the intersection points of the coordinates of the two aircraft and the straight line are respectively... Define q i,j The cross-determination coefficient between the two aircraft is calculated using the following formula:

[0111]

[0112] If p i,j =1, which means that the two aircraft are flying along the same route; if q i,j =1 indicates that the two aircraft are flying along different routes and intersecting, at which point the speed at which they intersect is c. i,j The calculation formula is as follows:

[0113]

[0114] Potential conflict time t between the two aircraft i,j The calculation formula is as follows:

[0115]

[0116] When c i,j >0, calculate the coordinates (x) of the intersection conflict point. i,j ,y i,j The formula is as follows:

[0117]

[0118] Then the conflict risk value of a certain spatiotemporal grid in the c-th time interval. The calculation formula is as follows:

[0119]

[0120] Among them, t i,j,c This represents the potential conflict time for aircraft i,j in the c-th time interval;

[0121] Furthermore, the total conflict risk value F of this spatiotemporal grid over M time intervals is... CR for:

[0122]

[0123] 4) Using the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), the K-Means clustering algorithm is used to classify the traffic complexity level of each spatiotemporal grid at different time periods;

[0124] The specific method is as follows:

[0125] 4.1) Standardize the traffic complexity assessment index data obtained in step 3), using the following formula:

[0126]

[0127] Where: Q is the traffic complexity assessment index data to be standardized, μ is the mean of the traffic complexity assessment index, and σ is the standard deviation of the traffic complexity assessment index.

[0128] 4.2) Using the traffic complexity assessment index of each time period of the standardized spatiotemporal grid as a sample, the traffic complexity level of each time period of the spatiotemporal grid is divided into four levels: normal, slightly complex, moderately complex and severely complex, and labeled with the numbers 1, 2, 3 and 4 respectively.

[0129] 5) Construct a terminal area traffic complexity level prediction model based on CNN (convolutional neural network). The input of this model is the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), and the output is the traffic complexity level of each spatiotemporal grid obtained in step 4). By learning the spatiotemporal evolution law of air traffic complexity through the terminal area traffic complexity level prediction model, the future traffic complexity level can be predicted based on historical traffic complexity index data.

[0130] The specific method is as follows:

[0131] 5.1) The traffic complexity assessment index data of each spatiotemporal grid obtained in step 4.1) and the traffic complexity level of the spatiotemporal grid obtained in step 4.2) constitute a sample set; each spatiotemporal grid corresponds to one sample, the feature data of which is the traffic complexity assessment index data of the spatiotemporal grid, and the label data is the traffic complexity level of the spatiotemporal grid; then 80% of the sample set is used as the training dataset, and the remaining 20% ​​is used as the test dataset.

[0132] 5.2) Input the samples from the above training dataset into the CNN for training. The CNN architecture is CNN1→CNN2→ReLU→Pooling→Softmax, where CNN1 and CNN2 are training layers used to select features, learn and train on the samples, and fit the traffic complexity assessment index data input in the previous step. CNN1 and CNN2 use three-dimensional convolution kernels, and the convolution operation function is Y=f(X×K+b), where X is the input K-dimensional convolution kernel, b is the bias, and f is the ReLU activation function. The first convolution layer is K1∈R 3×3×3×32 Output Y1∈R m×n×r×32 The second convolutional layer is K2∈R 5×5×5×64 Output Y2∈R m×n×r×64 Then, a max pooling operation is performed, with a pooling kernel S = 2 × 2 × 2 and a stride of 2. The first pooling layer is Y1 → Y′1, and the second pooling layer is Y2 → Y′2. Finally, the training results are classified after passing through a Softmax layer.

[0133] 5.3) Backpropagation is performed on the above CNN using the mean squared error loss function. During this process, relevant parameters in the neurons are updated. The cross-entropy loss function is used to measure the difference between the predicted probability and the true label, and the formula is as follows: Use the Adam optimizer to update network parameters to minimize the cross-entropy loss function. The training iteration termination condition is determined by calculating whether the loss result reaches the preset value.

[0134] 5.4) When the iteration termination condition is met, the CNN training is completed, thus obtaining the terminal area traffic complexity level prediction model. Then, the samples in the above test dataset are input into the prediction model to obtain the traffic complexity level prediction value.

[0135] 5.5) Compare the predicted traffic complexity level obtained in the previous step with the actual traffic complexity level to calculate the prediction accuracy, as shown in the following formula:

[0136]

[0137] Where N is the total number of samples in the sample set, Y i p Y is the predicted traffic complexity level for the i-th sample. i a Let be the actual value of the traffic complexity level for the i-th sample.

[0138] Here are some application examples of the data-driven terminal area traffic complexity assessment and prediction method provided by this invention:

[0139] (1) Collect the latitude and longitude coordinates, altitude, speed, acceleration and heading data of each track point in the actual flight track data of the three airports, John F. Kennedy International Airport, LaGuardia Airport and Newark International Airport, located in the New York terminal area from March 26, 2023 to April 25, 2023.

[0140] (2) Convert the latitude and longitude coordinates of all track points collected in step (1) into rectangular coordinates, as shown in Table 1; then detect abnormal track points and delete them; finally, use linear interpolation to fill in the missing track points at 1-second intervals. The results of partially filled track points are shown in Table 1. Figure 2 As shown.

[0141] Table 1. Sample table of coordinate system transformation

[0142]

[0143] (3) Divide the New York terminal area airspace into spatial grids. The preprocessed tracks obtained in step (1) are assigned to their respective spatial grids according to the spatial location of each track point. Then, the track data of each spatial grid is divided into different time intervals at 10-minute intervals, thereby assigning each track point to its respective spatiotemporal grid.

[0144] (4) Calculate the macro-level traffic complexity assessment indicators for each spatiotemporal grid: average speed, average acceleration, speed standard deviation, and traffic flow, as well as the micro-level traffic complexity assessment indicator, namely, conflict risk value. Some statistical results are shown in Table 2. Further time-series plots of each assessment indicator are drawn, such as... Figure 3 As shown.

[0145] Table 2. Sample Table of Complexity Indicators

[0146]

[0147] (5) Standardize the data of each evaluation index collected in the previous step, and then use the standardized traffic complexity evaluation indexes of each spatiotemporal grid at different time periods as samples. Use the K-Means clustering algorithm to divide them into 1-4 levels, where level 1 represents normal, level 2 represents mild complexity, level 3 represents moderate complexity, and level 4 represents severe complexity. Identify the traffic complexity level of each spatiotemporal grid at different time periods. The classification results are as follows: Figure 4 As shown.

[0148] (6) Using traffic complexity assessment index data as features and traffic complexity level as labels, sample data for each spatiotemporal grid is constructed. 80% of these samples are input into the CNN, and the remaining 20% ​​are used for testing. The prediction results are as follows: Figure 5 As shown, the prediction accuracy reached 87%.

[0149] Note: Contents not described in detail in this specification are prior art known to those skilled in the art. The solution of this invention can be implemented by combining the methods described in the specification with the art itself.

Claims

1. A data-driven method for assessing and predicting traffic complexity in terminal areas, characterized in that: The data-driven terminal area traffic complexity assessment and prediction method includes the following steps performed in sequence: 1) Collect the actual arrival and departure flight trajectory data of the terminal area to be tested; then convert the trajectory data from the latitude and longitude coordinate system to the rectangular coordinate system; finally clean the abnormal data in the trajectory data and fill in the missing data, thereby preprocessing the trajectory data and obtaining the preprocessed trajectory data. 2) Divide the terminal area airspace into a four-dimensional spatiotemporal grid, and then use the preprocessed track data obtained in step 1) to allocate each track point data to the corresponding four-dimensional spatiotemporal grid according to its spatial location and time period, so as to obtain the track data of each spatiotemporal grid. 3) Using the trajectory data of each spatiotemporal grid obtained in step 2), calculate the macro and micro assessment index data of traffic complexity; The specific method for calculating the micro-level assessment index of traffic complexity is as follows: In a plane, the two-dimensional coordinates of the two aircraft are respectively and Their speeds are respectively and The vector connecting the spatial positions of the two aircraft is ,definition The determination coefficient for two aircraft flying on the same route is calculated using the following formula: (3)； Based on the spatial coordinates and velocity vectors of the two aircraft, the following two-dimensional linear equation can be established: (4)； in These represent the aircraft at a certain moment. and aircraft Location spatial coordinates, These represent the velocity vectors of the two aircraft respectively. shaft and The ratio of the axial components, They represent aircraft and aircraft exist The intercept of the axis; If a solution exists when we solve the equations of the two lines mentioned above, then we can determine the intersection point of the lines. The line vectors connecting the intersection points of the coordinates of the two aircraft and the straight line are respectively ,definition The cross-determination coefficient between the two aircraft is calculated using the following formula: (5)； like This indicates that the two aircraft are flying along the same route; if This indicates that two aircraft are flying along different routes and intersecting, at which point the speed at which the two aircraft intersect is... The calculation formula is as follows: (6)； Potential conflict time between the two aircraft The calculation formula is as follows: (7)； when Calculate the coordinates of the intersection conflict point. The formula is as follows: (8)； Then the conflict risk value of a certain spatiotemporal grid in the c-th time interval. The calculation formula is as follows: (9)； in, Indicates aircraft In the Potential conflict times under each time interval; Furthermore, the total conflict risk value of this spatiotemporal grid over M time intervals is... for: (10)； 4) Using the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), the K-Means clustering algorithm is used to classify the traffic complexity level of each spatiotemporal grid at different time periods; 5) Construct a terminal area traffic complexity level prediction model based on CNN. The input of this model is the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), and the output is the traffic complexity level of each spatiotemporal grid obtained in step 4). By learning the spatiotemporal evolution law of air traffic complexity through the terminal area traffic complexity level prediction model, the future traffic complexity level can be predicted based on historical traffic complexity index data.

2. The data-driven terminal area traffic complexity assessment and prediction method according to claim 1, characterized in that: In step 1), the actual arrival and departure flight track data of the terminal area to be tested are collected; then the track data is transformed from the latitude and longitude coordinate system to the rectangular coordinate system; finally, abnormal data in the track data is cleaned and missing data is filled in. The specific method for preprocessing the track data and obtaining the preprocessed track data is as follows: 1.1) Track data acquisition Collect the latitude and longitude coordinates, altitude, speed, acceleration, and heading data of each track point in the actual arrival and departure flight track data of the terminal area to be tested; 1.2) Coordinate Transformation Convert the latitude and longitude coordinates of all track points collected in step 1.1) into Cartesian coordinates, using the following formula: (1)； (2)； Where a is the semi-major axis of the reference ellipsoid, b is the semi-minor axis of the reference ellipsoid, e is the first eccentricity, and e' is the second eccentricity, representing the latitude and longitude of the center point of the zero meridian. As the origin of the projection, The coordinates of the midpoint of the selected area. These are the longitude and latitude coordinates of the track point, respectively. These are the Mercator projection coordinates of the transformed track points. These are the Cartesian coordinates of the transformed track points; 1.3) Deletion of abnormal waypoints Calculate the angle between any three consecutive track points on each track after the coordinate transformation obtained in step 1.2). If the angle is acute, it indicates that the middle track point among the three track points is an abnormal track point and it is deleted. 1.4) Filling in missing waypoints Linear interpolation is used to interpolate each track in step 1.3) at 1-second intervals to fill in missing track points; finally, the preprocessed track data is obtained.

3. The data-driven terminal area traffic complexity assessment and prediction method according to claim 1, characterized in that: In step 2), the terminal area airspace is divided into a four-dimensional spatiotemporal grid. Then, using the preprocessed track data obtained in step 1), each track point data is assigned to the corresponding four-dimensional spatiotemporal grid according to its spatial location and time period. The specific method for obtaining the track data of each spatiotemporal grid is as follows: 2.1) Spatial grid division The terminal area airspace is in a rectangular coordinate system Divide into equal parts along the axial direction This divides the horizontal projection plane of the terminal area airspace into sections.

1. Two adjacent horizontal grids; then, in the vertical direction, divide into equal parts by unit length. This divides the terminal area airspace into segments. The three-dimensional spatial grid is used, and each three-dimensional spatial grid represents a specific spatial region; then, the track points in the preprocessed track obtained in step 1) are assigned to their respective spatial grids according to their spatial positions; 2.2) Time grid partitioning by For each time interval, the track point data of each spatial grid obtained in step 2.2) are assigned to the corresponding time interval according to their time, thereby dividing the terminal area airspace into a four-dimensional spatiotemporal grid and obtaining the track data of each spatiotemporal grid.

4. The data-driven terminal area traffic complexity assessment and prediction method according to claim 3, characterized in that: In step 3), the specific method for calculating the macro-level assessment index of traffic complexity is as follows: Using the track data of each spatiotemporal grid obtained in step 2.2), calculate the average velocity of all track points within any spatiotemporal grid, and use this average velocity as the average velocity of that spatiotemporal grid, denoted as . Calculate the average acceleration of all track points within any spatiotemporal grid, and use this as the average acceleration of that spatiotemporal grid, denoted as . ; Calculate the standard deviation of the velocities of all trackpoints within any spatiotemporal grid, and use this as the velocity standard deviation of that spatiotemporal grid, denoted as . ; Calculate the number of aircraft passing through any spatiotemporal grid, which is taken as the traffic flow of that spatiotemporal grid, denoted as . .

5. The data-driven terminal area traffic complexity assessment and prediction method according to claim 1, characterized in that: In step 4), the specific method for using the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3) to classify the traffic complexity level of each spatiotemporal grid for different time periods using the K-Means clustering algorithm is as follows: 4.1) Standardize the traffic complexity assessment index data obtained in step 3), using the following formula: （11）； Where: Q represents a traffic complexity assessment index data to be standardized. Let be the mean of a certain traffic complexity assessment index. Let be the standard deviation of a certain traffic complexity assessment index; 4.2) Using the traffic complexity assessment index of each spatiotemporal grid at different time periods after standardization as samples, the traffic complexity level of each spatiotemporal grid at different time periods is divided into four levels: normal, slightly complex, moderately complex and severely complex, and labeled with the numbers 1, 2, 3 and 4 respectively.

6. The data-driven terminal area traffic complexity assessment and prediction method according to claim 1, characterized in that: In step 5), the terminal area traffic complexity level prediction model is constructed based on CNN. The input of this model is the traffic complexity assessment index data of each spatiotemporal grid obtained in step 3), and the output is the traffic complexity level of each spatiotemporal grid obtained in step 4). The specific method for predicting the future traffic complexity level based on historical traffic complexity index data by learning the spatiotemporal evolution law of air traffic complexity through the terminal area traffic complexity level prediction model is as follows: 5.1) The traffic complexity assessment index data of each spatiotemporal grid obtained in step 4.1) and the traffic complexity level of the spatiotemporal grid obtained in step 4.2) constitute a sample set; each spatiotemporal grid corresponds to one sample, the feature data of which is the traffic complexity assessment index data of the spatiotemporal grid, and the label data is the traffic complexity level of the spatiotemporal grid; then 80% of the sample set is used as the training dataset, and the remaining 20% ​​is used as the test dataset. 5.2) Input the samples from the above training dataset into the CNN for training. The CNN architecture is CNN1. CNN2 Relu Softmax, where CNN1 and CNN2 are training layers used to select features, learn and train on samples, and fit the traffic complexity assessment index data input in the previous step. CNN1 and CNN2 use three-dimensional convolution kernels, and the convolution operation function is... ,in For input 3D convolution kernel, For bias, The activation function is ReLU, where the first convolutional layer is... Output The second convolutional layer is Output Then perform max pooling operation, pooling kernel The step size is 2, where the first pooling layer is The second pooling layer is Finally, the training results are classified after passing through the Softmax layer; 5.3) The mean squared error loss function is used to backpropagate the above CNN. During this process, the relevant parameters in the neurons are updated. The cross-entropy loss function is used to measure the difference between the predicted probability and the true label. The formula is as follows: The Adam optimizer is used to update network parameters to minimize the cross-entropy loss function. The training iteration termination condition is determined by calculating whether the loss result reaches the preset value. 5.4) When the iteration termination condition is met, the CNN training is completed, thus obtaining the terminal area traffic complexity level prediction model. Then, the samples in the above test dataset are input into the prediction model to obtain the traffic complexity level prediction value. 5.5) Compare the predicted traffic complexity level obtained in the previous step with the actual traffic complexity level to calculate the prediction accuracy, using the following formula: (12)； Where N is the total number of samples in the sample set. Let i be the predicted traffic complexity level for the i-th sample. Let be the actual value of the traffic complexity level for the i-th sample.

Images