Multi-aircraft track credibility prediction method and device based on machine learning

Abstract

According to the multi-aircraft track credibility prediction method and device based on machine learning, track credibility prediction can be rapidly carried out on future surrounding aircrafts, and powerful support is provided for various tasks such as arrival and departure sorting, conflict detection, flight flow management and arrival time estimation in the field of air traffic flow management. The method comprises the following steps: (1) carrying out data preprocessing, constructing a multi-aircraft track data set, and converting data of a plurality of complex scenes into a fixed-length track vector sequence; (2) a multi-aircraft track credible prediction model is constructed, a framework is constructed based on spatial-temporal correlation of multi-aircraft tracks, the framework comprises a spatial feature extraction module and a long short-term memory (LSTM) network module, the spatial feature extraction module enables adjacent agents to interact in a modeling space, and the LSTM module captures a time dependency relationship of a 4D track of a single aircraft; and (3) training by using historical data, and carrying out credibility prediction on the multi-aircraft track by using the multi-aircraft track credibility prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent civil aviation air traffic management, and in particular to a reliable prediction method for multi-aircraft trajectories based on machine learning, and also to a reliable prediction device for multi-aircraft trajectories based on machine learning. Background Art

[0002] Whether it is the current sector-based operation mode or the future trajectory-based operation (TBO) mode, reliable prediction of aircraft trajectory is the key to the efficient operation of these two modes. At the same time, accurate reliable trajectory prediction is a key technology for various applications in the field of air traffic management (ATM), such as approach and departure sequencing, conflict detection, flight flow management, and arrival time estimation. It can greatly reduce the uncertainty of future aircraft flights and improve the predictability of air traffic.

[0003] Due to airspace congestion, weather, temporary military activities, airspace restrictions and other reasons, the actual flight process of the aircraft sometimes does not follow the pre-planned route. This reduces the predictability of air traffic, increases the difficulty of air traffic management, and seriously hinders the efficient operation of the air traffic management system. During the flight, the aircraft will be affected by other adjacent aircraft that overlap at the same time. For example, controllers sometimes specify that the aircraft deviate from the planned route to alleviate airspace congestion. This trajectory deviation cannot be predicted by observing a single aircraft alone, and it is necessary to consider other aircraft around it that are most likely to affect it.

[0004] The difficulties in reliable prediction of multi-aircraft trajectories are: (1) it is difficult to capture the interactions between agents in a dynamic environment. For example, two agents can interact in the current time step but not in the future; (2) the action space of multiple agents is significantly larger than that of a single agent. For example, in each step, the action of one agent has an impact on the actions of its neighboring agents; and 3) temporal correlation is coupled with spatial interaction. For example, the time series motion of the current agent needs to consider the motion of all its neighbors. Summary of the invention

[0005] In order to overcome the defects of the prior art, the technical problem to be solved by the present invention is to provide a reliable prediction method for multi-aircraft trajectories based on machine learning, which can quickly make reliable predictions on the trajectories of future surrounding aircraft, and provide strong support for various tasks in the field of air traffic flow management, such as approach and departure sorting, conflict detection, flight flow management, arrival time estimation, etc.

[0006] The technical solution of the present invention is: the reliable prediction method of multiple aircraft tracks based on machine learning comprises the following steps:

[0007] (1) Perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into fixed-length track vector sequences to adapt to the input format of the machine learning model;

[0008] (2) Construct a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of multi-aircraft trajectories, a framework is constructed. The framework includes a spatial feature extraction module and a long short-term memory network (LSTM) module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft.

[0009] (3) Use historical data for training and save the trained model data. Then use the multi-aircraft track credible prediction model to make credible predictions on the tracks of multiple aircraft.

[0010] Based on the spatiotemporal correlation of multi-aircraft trajectories, the present invention constructs a credible prediction model for multi-aircraft trajectories, which can capture the action mode of a single aircraft and the interaction of intelligent agents in crowded scenes in a data-driven manner, and has strong generalization ability. It can quickly make credible predictions of multi-aircraft trajectories with a certain degree of confidence and confidence interval, meet real-time requirements, and provide strong support for various tasks in the field of air traffic flow management, such as approach and departure sorting, conflict detection, flight flow management, and arrival time estimation.

[0011] A multi-aircraft track credible prediction device based on machine learning is also provided, which comprises:

[0012] A data preprocessing module, which is configured to perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into a fixed-length track vector sequence to fit the input format of the machine learning model;

[0013] A model building module configured to build a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of the trajectories of multiple aircraft, a framework is built, which includes a spatial feature extraction module and a long short-term memory network LSTM module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft.

[0014] The training module is configured to use historical data for training, save the trained model data, and then use the multi-aircraft track credible prediction model to make credible predictions on the multi-aircraft tracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Shown is an overall flow chart of the reliable prediction method for multiple aircraft trajectories based on machine learning according to the present invention.

[0016] Figure 2 Shown is a flowchart of step (2) of the method for reliable prediction of multi-aircraft trajectories based on machine learning according to the present invention.

[0017] Figure 3 The flowchart shown is a spatial feature extraction module of step (2) of the method for reliable prediction of multi-aircraft trajectories based on machine learning according to the present invention, which enables the interaction of adjacent intelligent agents in the modeling space. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] In order to make the description of the present disclosure more detailed and complete, the following provides an illustrative description of the implementation and specific embodiments of the present invention; however, this is not the only form of implementing or using the specific embodiments of the present invention. The implementation covers the features of multiple specific embodiments and the method steps and sequences used to construct and operate these specific embodiments. However, other specific embodiments can also be used to achieve the same or equivalent functions and step sequences.

[0020] like Figure 1 As shown, the multi-aircraft trajectory reliable prediction method based on machine learning includes the following steps:

[0021] (1) Perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into fixed-length track vector sequences to adapt to the input format of the machine learning model;

[0022] (2) Construct a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of multi-aircraft trajectories, a framework is constructed. The framework includes a spatial feature extraction module and a long short-term memory network (LSTM) module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft.

[0023] (3) Use historical data for training and save the trained model data. Then use the multi-aircraft track credible prediction model to make credible predictions on the tracks of multiple aircraft.

[0024] Based on the spatiotemporal correlation of multi-aircraft trajectories, the present invention constructs a credible prediction model for multi-aircraft trajectories, which can capture the action mode of a single aircraft and the interaction of intelligent agents in crowded scenes in a data-driven manner, and has strong generalization ability. It can quickly make credible predictions of multi-aircraft trajectories with a certain degree of confidence and confidence interval, meet real-time requirements, and provide strong support for various tasks in the field of air traffic flow management, such as approach and departure sorting, conflict detection, flight flow management, and arrival time estimation.

[0025] Preferably, the step (1) comprises the following sub-steps:

[0026] (1.1) Construct M 4D track data sets of multiple aircraft, where M is the number of track samples. The track data of multiple aircraft in the same scene are divided into track vector sequences with a time interval of t, a length of L, and a maximum number of aircraft N. The track vectors include: aircraft id, longitude, latitude, heading, and ground speed;

[0027] (1.2) Normalize the data of all track samples and scale them to the range of (0, 1) using the Min-Max normalization method;

[0028] (1.3) Randomly select M track samples to construct M multi-aircraft 4D track datasets.

[0029] Preferably, the spatial feature extraction module of step (2) enables the interaction of adjacent agents in the modeling space, including:

[0030] Set the spatial grid size to indicate that within a certain spatial distance, the aircraft adjusts its speed and heading due to the situation of surrounding aircraft, and the number of spatial grids N0 indicates the accuracy of the aircraft's perception of the situation of surrounding aircraft. For each aircraft, the spatial grids to which other aircraft belong are calculated according to the position of other aircraft relative to the aircraft using the following indicator function:

[0031]

[0032] When the position of aircraft j is within the (m,n) grid of aircraft i, The value is 1. At each time t, the hidden layer state h of LSTM i t Capture the potential representation of aircraft i; construct the Social tensor It is used to share the representation between adjacent aircraft to model the spatial interaction relationship between aircraft at the current moment. Given the hidden state dimension D and the number of spatial grids N0, is a tensor of size N0×N0×D, calculated as follows:

[0033]

[0034] in is the hidden layer state of aircraft j at time t-1, is the indicator function of aircraft j relative to aircraft i in the spatial grid (m,n).

[0035] Preferably, the LSTM module of step (2) captures the temporal dependency of the 4D trajectory of a single aircraft, including:

[0036] The track vector of length 5 obtained in step (1) is input into an embedding layer to obtain the track feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula:

[0037]

[0038] Where W e is the weight of the fully connected layer. The obtained Social tensor is input into an embedding layer to obtain the spatial feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula:

[0039]

[0040] Where W e is the weight of the fully connected layer. The above track feature vector and spatial feature vector are concatenated to obtain a vector of length 128 as the input vector of the LSTM module. In order to Bayesianize this module, a dropout-mask is set for each aircraft and each timestamp to randomly discard a part of the neurons representing the input vector. The LSTM module includes a layer of LSTM units, a dropout applied to the LSTM parameters, and a fully connected output layer. All aircraft share one LSTM module to capture the time correlation of each track, and finally obtain the prediction result represented by probability. One cycle of LSTM is expressed by the formula:

[0041]

[0042] in, They represent the hidden layer state and cell state of LSTM at time step t, and the hidden layer state and cell state of LSTM at time step t-1, respectively.

[0043] Preferably, in the training phase of step (3), a large number of track data sets including multiple aircraft are prepared and divided into a training set, a validation set and a test set according to a ratio of 5:1:1; first, the model is trained in the training set, and one piece of data is trained in each round. At the beginning, due to the large range of the solution space, a large learning rate is adopted, and then the learning rate decay method is adopted to gradually reduce the learning rate to obtain stable training; the validation set is used to forward propagate the model every several rounds. If the loss of the model can reach a certain threshold, the training is considered successful and the current neural network parameters are saved; the scalability of the model is retained, and the model is iteratively upgraded as historical data is enriched and accumulated.

[0044] Preferably, in the testing phase of step (3), the saved neural network model is used to perform testing in the test set, the neural network is forward propagated n times, the mean and variance of the n results are calculated, and a predicted trajectory with a certain confidence level and confidence interval is quickly obtained.

[0045] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes the steps of the above-mentioned embodiment method, and the storage medium can be: ROM / RAM, disk, CD, memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a reliable prediction device for multi-aircraft tracks based on machine learning, which is usually represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0046] A data preprocessing module, which is configured to perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into a fixed-length track vector sequence to fit the input format of the machine learning model;

[0047] A model building module configured to build a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of the trajectories of multiple aircraft, a framework is built, which includes a spatial feature extraction module and a long short-term memory network LSTM module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft.

[0048] The training module is configured to use historical data for training, save the trained model data, and then use the multi-aircraft track credible prediction model to make credible predictions on the multi-aircraft tracks.

[0049] Preferably, the spatial feature extraction module of the model building module enables the interaction of adjacent agents in the modeling space, including:

[0050] Set the spatial grid size to indicate that within a certain spatial distance, the aircraft adjusts its speed and heading due to the situation of surrounding aircraft, and the number of spatial grids N0 indicates the accuracy of the aircraft's perception of the situation of surrounding aircraft. For each aircraft, the spatial grids to which other aircraft belong are calculated according to the position of other aircraft relative to the aircraft using the following indicator function:

[0051]

[0052] When the position of aircraft j is within the (m,n) grid of aircraft i, The value is 1. At each time t, the hidden layer state of LSTM Capture the potential representation of aircraft i; construct the Social tensor It is used to share the representation between adjacent aircraft to model the spatial interaction relationship between aircraft at the current moment. Given the hidden state dimension D and the number of spatial grids N0, is a tensor of size N0×N0×D, calculated as follows:

[0053]

[0054] in is the hidden layer state of aircraft j at time t-1.

[0055] Preferably, the LSTM module of step (2) captures the temporal dependency of the 4D trajectory of a single aircraft, including:

[0056] The track vector of length 5 obtained in step (1) is input into an embedding layer to obtain the track feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula:

[0057]

[0058] Where W e is the weight of the fully connected layer. The obtained Social tensor is input into an embedding layer to obtain the spatial feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula:

[0059]

[0060] Where W eis the weight of the fully connected layer. The above track feature vector and spatial feature vector are concatenated to obtain a vector of length 128 as the input vector of the LSTM module. In order to Bayesianize this module, a dropout-mask is set for each aircraft and each timestamp to randomly discard a part of the neurons representing the input vector. The LSTM module includes a layer of LSTM units, a dropout applied to the LSTM parameters, and a fully connected output layer. All aircraft share one LSTM module to capture the time correlation of each track, and finally obtain the prediction result represented by probability. One cycle of LSTM is expressed by the formula:

[0061]

[0062] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the protection scope of the technical solution of the present invention.

Claims

1. A reliable prediction method for multi-aircraft tracks based on machine learning, characterized by: It includes the following steps: (1) Perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into fixed-length track vector sequences to adapt to the input format of the machine learning model; (2) Construct a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of multi-aircraft trajectories, a framework is constructed. The framework includes a spatial feature extraction module and a long short-term memory network (LSTM) module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft. (3) Use historical data for training and save the trained model data. Then use the multi-aircraft track credible prediction model to make credible predictions on the tracks of multiple aircraft.

2. The method for reliable prediction of multi-aircraft tracks based on machine learning according to claim 1, characterized in that: The step (1) comprises the following sub-steps: (1.1) Construct M 4D track data sets of multiple aircraft, where M is the number of track samples. The track data of multiple aircraft in the same scene are divided into track vector sequences with a time interval of t, a length of L, and a maximum number of aircraft N. The track vectors include: aircraft id, longitude, latitude, heading, and ground speed; (1.2) Normalize the data of all track samples and scale them to the range of (0, 1) using the Min-Max normalization method; (1.3) Randomly select M track samples to construct M multi-aircraft 4D track datasets.

3. The method for reliable prediction of multi-aircraft tracks based on machine learning according to claim 2, characterized in that: The spatial feature extraction module of step (2) enables the interaction of adjacent intelligent agents in the modeling space, including: Set the spatial grid size to indicate that within a certain spatial distance, the aircraft adjusts its speed and heading due to the situation of surrounding aircraft, and the number of spatial grids N0 indicates the accuracy of the aircraft's perception of the situation of surrounding aircraft. For each aircraft, the spatial grids to which other aircraft belong are calculated according to the position of other aircraft relative to the aircraft using the following indicator function: When the position of aircraft j is within the (m,n) grid of aircraft i, The value is 1. At each time t, the hidden layer state of LSTM Capture the potential representation of aircraft i; construct the Social tensor It is used to share the representation between adjacent aircraft to model the spatial interaction relationship between aircraft at the current moment. Given the hidden state dimension D and the number of spatial grids N0, is a tensor of size N0×N0×D, calculated as follows: in is the hidden layer state of aircraft j at time t-1, is the indicator function of aircraft j relative to aircraft i in the spatial grid (m,n).

4. The method for reliable prediction of multi-aircraft tracks based on machine learning according to claim 3, characterized in that: The LSTM module of step (2) captures the temporal dependency of the 4D trajectory of a single aircraft, including: The track vector of length 5 obtained in step (1) is input into an embedding layer to obtain the track feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula: Where W e is the weight of the fully connected layer. The obtained Social tensor is input into an embedding layer to obtain the spatial feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula: Where W e is the weight of the fully connected layer. The above track feature vector and spatial feature vector are concatenated to obtain a vector of length 128 as the input vector of the LSTM module. In order to Bayesianize this module, a dropout-mask is set for each aircraft and each timestamp to randomly discard a part of the neurons representing the input vector. The LSTM module includes a layer of LSTM units, a dropout applied to the LSTM parameters, and a fully connected output layer. All aircraft share one LSTM module to capture the time correlation of each track, and finally obtain the prediction result represented by probability. One cycle of LSTM is expressed by the formula: in, They represent the hidden layer state and cell state of LSTM at time step t, and the hidden layer state and cell state of LSTM at time step t-1, respectively.

5. The method for reliable prediction of multi-aircraft tracks based on machine learning according to claim 4, characterized in that: In the training phase of step (3), a large number of track data sets including multiple aircraft are prepared and divided into a training set, a validation set and a test set according to a ratio of 5:1:1; first, the model is trained in the training set, and one piece of data is trained in each round. At the beginning, due to the large range of the solution space, a large learning rate is adopted, and then the learning rate decay method is adopted to gradually reduce the learning rate to obtain stable training; the validation set is used to forward propagate the model every several rounds. If the loss of the model can reach a certain threshold, the training is considered successful and the current neural network parameters are saved; the scalability of the model is retained, and the model is iteratively upgraded as the historical data is enriched and accumulated.

6. The method for reliable prediction of multi-aircraft tracks based on machine learning according to claim 5, characterized in that: In the test phase of step (3), the saved neural network model is used to test in the test set, the neural network is forward propagated n times, the mean and variance of the n results are calculated, and the prediction trajectory with a certain confidence level and confidence interval is quickly obtained.

7. A reliable prediction device for multi-aircraft flight paths based on machine learning, characterized in that: It includes: A data preprocessing module, which is configured to perform data preprocessing and construct a multi-aircraft track dataset, converting data from multiple complex scenarios into a fixed-length track vector sequence to fit the input format of the machine learning model; A model building module configured to build a reliable prediction model for multi-aircraft trajectories. Based on the spatiotemporal correlation of the trajectories of multiple aircraft, a framework is built, which includes a spatial feature extraction module and a long short-term memory network LSTM module. The spatial feature extraction module enables the interaction of adjacent intelligent agents in the modeling space, and the LSTM module captures the temporal dependency of the 4D trajectories of single aircraft. The training module is configured to use historical data for training, save the trained model data, and then use the multi-aircraft track credible prediction model to make credible predictions on the multi-aircraft tracks.

8. The device for reliable prediction of multi-aircraft tracks based on machine learning according to claim 7, characterized in that: The data preprocessing module performs the following steps: (1.1) Construct M 4D track data sets of multiple aircraft, where M is the number of track samples. The track data of multiple aircraft in the same scene are divided into track vector sequences with a time interval of t, a length of L, and a maximum number of aircraft N. The track vectors include: aircraft id, longitude, latitude, heading, and ground speed; (1.2) Normalize the data of all track samples and scale them to the range of (0, 1) using the Min-Max normalization method; (1.3) Randomly select M track samples to construct M multi-aircraft 4D track datasets.

9. The device for reliable prediction of multi-aircraft tracks based on machine learning according to claim 8, characterized in that: The spatial feature extraction module of the model building module enables the interaction of adjacent intelligent agents in the modeling space, including: Set the spatial grid size to indicate that within a certain spatial distance, the aircraft adjusts its speed and heading due to the situation of surrounding aircraft, and the number of spatial grids N0 indicates the accuracy of the aircraft's perception of the situation of surrounding aircraft. For each aircraft, the spatial grids to which other aircraft belong are calculated according to the position of other aircraft relative to the aircraft using the following indicator function: When the position of aircraft j is within the (m,n) grid of aircraft i, The value is 1. At each time t, the hidden layer state of LSTM Capture the potential representation of aircraft i; construct the Social tensor It is used to share the representation between adjacent aircraft to model the spatial interaction relationship between aircraft at the current moment. Given the hidden state dimension D and the number of spatial grids N0, is a tensor of size N0×N0×D, calculated as follows: in is the hidden layer state of aircraft j at time t-1.

10. The device for reliable prediction of multi-aircraft tracks based on machine learning according to claim 9, characterized in that: The LSTM module of step (2) captures the temporal dependency of the 4D trajectory of a single aircraft, including: The track vector of length 5 obtained in step (1) is input into an embedding layer to obtain the track feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula: Where W e is the weight of the fully connected layer. The obtained Social tensor is input into an embedding layer to obtain the spatial feature vector. The embedding layer includes a fully connected layer and a ReLU nonlinear layer, which is expressed by the following formula: Where W e is the weight of the fully connected layer. The above track feature vector and spatial feature vector are concatenated to obtain a vector of length 128 as the input vector of the LSTM module. In order to Bayesianize this module, a dropout-mask is set for each aircraft and each timestamp to randomly discard a part of the neurons representing the input vector. The LSTM module includes a layer of LSTM units, a dropout applied to the LSTM parameters, and a fully connected output layer. All aircraft share one LSTM module to capture the time correlation of each track, and finally obtain the prediction result represented by probability. One cycle of LSTM is expressed by the formula:

Images