Aircraft tracking method, device and equipment based on multi-source sensor data

Abstract

The invention relates to the technical field of communication, and provides an aircraft tracking method, device and equipment based on multi-source sensor data, and the method comprises the steps: obtaining to-be-predicted sensor data sets corresponding to a target aircraft at a plurality of moments, inputting the plurality of to-be-predicted sensor data sets into a trajectory tracking model, the trajectory tracking model is obtained through training according to a training set, and the training set comprises training sensor data sets corresponding to multiple moments and corresponding real flight trajectory points, so that the training sensor data sets and the real flight trajectory points can be spatially correlated, and the real flight trajectory points can be spatially correlated; therefore, the model can effectively track the flight path through the sensor data. The trajectory tracking model comprises a feature embedding layer, a bidirectional gating circulation unit encoder and a gating circulation unit decoder, the bidirectional gating circulation unit encoder can obtain rich time sequence feature information at multiple moments, and the accuracy of flight trajectory tracking is improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular, to an aircraft tracking method, device, and equipment based on multi-source sensor data. Background Art

[0002] The core tasks of air traffic management include: airspace planning, flight interval management, flight scheduling, etc., all of which rely on accurate prediction of aircraft trajectories. Existing traditional methods mainly rely on the Automatic Dependent Surveillance-Broadcast (ADS-B) system to obtain flight trajectory data of aircraft. However, in the process of obtaining flight trajectory data through the ADS-B system, it is vulnerable to external environmental interference. For example, the signal of the ADS-B system may be interrupted, lagged, or maliciously attacked, resulting in missing or untrustworthy aircraft trajectory data, threatening the safety and stability of air traffic management.

[0003] In recent years, for the above problems, it is possible to collect sensor data of aircraft flight through a ground distributed sensor network, and realize the tracking of aircraft trajectories through the sensor data. However, for the sensor data collected by sensors, there are problems such as heterogeneity, non-linearity, clock drift, noise interference, and inaccurate position. Moreover, traditional physical prediction models, such as Kalman filtering and probability models, rely too much on prior information and lack dynamic adaptability, while single deep learning-based prediction models, such as models for time series processing like LSTM, lack in-depth research on tracking aircraft trajectories based on sensor data.

[0004] Based on this, the above physical prediction models and single deep learning-based prediction models cannot effectively track aircraft trajectories based on sensor data. Therefore, how to realize the tracking of aircraft trajectories only based on the sensor data collected by a ground distributed sensor network is an urgent problem to be solved currently. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide an aircraft tracking method, device, and equipment based on multi-source sensor data.

[0006] In a first aspect, an embodiment of the present invention provides an aircraft tracking method based on multi-source sensor data, the method including:

[0007] Obtain a to-be-predicted sensor data set corresponding to a target aircraft at multiple moments respectively, where the to-be-predicted sensor data set includes to-be-predicted data of multiple different sensors;

[0008] Input multiple of the to-be-predicted sensor data sets into the trained trajectory tracking model to obtain the flight trajectory tracking result corresponding to the target aircraft;

[0009] Among them, the trajectory tracking model is obtained by training according to a training set. The training set includes: training sensor data sets corresponding to multiple moments, and the true flight trajectory points corresponding to the training sensor data sets. The training sensor data sets include: training sensor data collected by multiple different sensors. The trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder.

[0010] In one embodiment, before inputting multiple of the to-be-predicted sensor data sets into the trained trajectory tracking model to obtain the flight trajectory tracking result corresponding to the target aircraft, it further includes:

[0011] Construct a training set according to the historical flight data of multiple aircraft and the sensor data collected by multiple sensors;

[0012] Input the training set into the initial trajectory tracking model, train the initial trajectory tracking model according to the teacher forcing strategy, and adjust the model parameters through a loss function until the model converges to obtain the trained trajectory tracking model.

[0013] In one embodiment, the constructing a training set according to the historical flight data of multiple aircraft and the sensor data collected by multiple sensors includes:

[0014] For multiple aircraft, collect each sensor data set of each aircraft at multiple moments. Among them, each sensor data set corresponding to each moment includes: sensor data collected by multiple different sensors;

[0015] Determine the flight coordinate position of each aircraft at multiple moments according to the historical flight data;

[0016] Construct a training set according to each sensor data set corresponding to multiple aircraft at multiple moments respectively and each flight coordinate position corresponding to each sensor data set.

[0017] In one embodiment, before constructing a training set according to each sensor data set corresponding to multiple aircraft at multiple moments respectively and each flight coordinate position corresponding to each sensor data set, it further includes:

[0018] Perform normalized data processing on each sensor data set corresponding to multiple aircraft at multiple moments respectively and each flight coordinate position corresponding to each sensor data set.

[0019] In one embodiment, inputting the training set into the initial trajectory tracking model, training the initial trajectory tracking model according to the teacher forcing strategy, and adjusting the model parameters through a loss function until the model converges to obtain the trained trajectory tracking model includes:

[0020] Input the training sensor data sets at N moments into the feature embedding layer to extract the high-dimensional feature vectors corresponding to the training sensor data sets at each moment;

[0021] Input the N high-dimensional feature vectors into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain hidden state features, where the hidden state features include N time series features related to the high-dimensional feature vectors at N moments;

[0022] After obtaining the hidden state features, according to the initial flight trajectory points, the hidden state features, and the true flight trajectory points, adopt the teacher forcing strategy and use the gated recurrent unit decoder to obtain the predicted flight trajectory points of the target aircraft at each moment;

[0023] Compare the predicted flight trajectory points with the true flight trajectory points, and adjust the model parameters according to the loss function until the model converges to obtain the trained trajectory tracking model.

[0024] In one embodiment, inputting the N high-dimensional feature vectors into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain hidden state features includes:

[0025] Input the first high-dimensional feature vector at the Mth moment, the first forward time series feature at the M - 1th moment, and the second forward time series feature at the M + 1th moment into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the time series feature at the Mth moment, where M = < N;

[0026] Determine the hidden state features according to the time series features at N moments.

[0027] In one embodiment, the bidirectional encoding processing includes forward encoding processing and backward encoding processing. Inputting the first high-dimensional feature vector at the Mth moment, the first forward time series feature at the M - 1th moment, and the second forward time series feature at the M + 1th moment into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the time series feature at the Mth moment includes:

[0028] Input the first high-dimensional feature vector and the first forward time series feature into the bidirectional gated recurrent unit encoder for forward encoding processing to obtain the forward time series feature at the Mth moment;

[0029] Input the first high-dimensional feature vector and the second forward time series feature into the bidirectional gated recurrent unit encoder for reverse encoding processing to obtain the reverse time series feature at the Mth moment;

[0030] Perform splicing processing based on the forward time series feature and the reverse time series feature to obtain the time series feature at the Mth moment.

[0031] In one embodiment, after obtaining the hidden state feature, according to the initial flight trajectory point, the hidden state feature, and the true flight trajectory point, adopting the teacher forcing strategy, through the gated recurrent unit decoder, obtaining the predicted flight trajectory points of the target aircraft at each moment includes:

[0032] When M is 0, input the initial flight trajectory point and the hidden state feature into the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment;

[0033] According to the teacher forcing strategy, determine the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point, and update the hidden state feature using the target decoding input feature to obtain the updated hidden state feature;

[0034] Input the target decoding input feature and the updated hidden state feature into the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment, and return to execute to determine the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point until the predicted flight trajectory points at N moments are obtained.

[0035] In a second aspect, an embodiment of the present invention provides an aircraft tracking device based on multi-source sensor data, including:

[0036] An acquisition module, configured to acquire a to-be-predicted sensor data set corresponding to a target aircraft at multiple moments, where the to-be-predicted sensor data set includes to-be-predicted data of multiple different sensors;

[0037] A flight trajectory tracking result acquisition module, configured to input multiple to-be-predicted sensor data sets into a trained trajectory tracking model to obtain a flight trajectory tracking result corresponding to the target aircraft;

[0038] Among them, the trajectory tracking model is obtained by training based on a training set, the training set includes: training sensor data sets corresponding to multiple moments, and true flight trajectory points corresponding to the training sensor data sets, the training sensor data sets include: training sensor data collected by multiple different sensors, and the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder.

[0039] In a third aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the aircraft tracking method based on multi-source sensor data described in the first aspect are implemented.

[0040] The technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:

[0041] An aircraft tracking method based on multi-source sensor data provided by an embodiment of the present invention, in this way, by obtaining a to-be-predicted sensor data set including to-be-predicted data of multiple different sensors corresponding to a target aircraft at multiple moments respectively, inputting the multiple to-be-predicted sensor data sets into a trained trajectory tracking model, and obtaining a flight trajectory tracking result corresponding to the target aircraft. The trajectory tracking model is obtained by training based on a training set. Since the training set includes training sensor data sets corresponding to multiple moments respectively, and true flight trajectory points corresponding to the training sensor data sets, spatial association between the training sensor data sets and the true flight trajectory points can be achieved, so that the trajectory tracking model can effectively implement flight trajectory tracking of the target aircraft through sensor data, and also solve the problem of relying on the ADS-B system to obtain prior information in the prior art. And the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder. Since the bidirectional gated recurrent unit encoder can obtain rich time series feature information at multiple moments, the accuracy of flight trajectory tracking of the target aircraft is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0044] Figure 1A schematic diagram of a scenario application provided by an embodiment of the present invention.

[0045] Figure 2 A schematic flowchart of a method for tracking an aircraft based on multi-source sensor data provided by an embodiment of the present invention;

[0046] Figure 3 A schematic structural diagram of a trajectory prediction model provided by an embodiment of the present invention;

[0047] Figure 4 A schematic structural diagram of a device for tracking an aircraft based on multi-source sensor data provided by an embodiment of the present invention. Detailed implementation manners

[0048] In order to more clearly understand the above objects, features and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0049] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present invention, rather than all the embodiments.

[0050] The core tasks of air traffic management include: airspace planning, flight interval management, flight scheduling, etc., all of which rely on the accurate prediction of aircraft trajectories. Refer to Figure 1 As shown, the existing traditional methods mainly rely on the Automatic Dependent Surveillance-Broadcast (ADS-B) system to obtain real-time trajectory data of aircraft flights. However, during the process of obtaining flight trajectory data through the ADS-B system, it is vulnerable to external environmental interference. For example, the signal of the ADS-B system may be interrupted, lagged or maliciously attacked, resulting in the loss or untrustworthiness of aircraft trajectory data, threatening the safety and stability of air traffic management.

[0051] In recent years, for the above problems, it is possible to collect sensor data of aircraft flights through a ground distributed sensor network and achieve the tracking of aircraft trajectories through this sensor data. However, for the sensor data collected by sensors, there are problems such as heterogeneity, non-linearity, clock drift, noise interference and inaccurate position, and traditional physical prediction models, such as Kalman filtering and probability models, rely too much on prior information and lack dynamic adaptability, while single deep learning-based prediction models, such as time series processing models like LSTM, lack in-depth research on tracking aircraft trajectories based on sensor data.

[0052] Based on this, the above physical prediction model and a single deep learning-based prediction model cannot effectively track the aircraft trajectory based on sensor data. Therefore, how to track the aircraft trajectory only based on the sensor data collected by the ground distributed sensor network is an urgent problem to be solved currently.

[0053] Therefore, the present invention provides an aircraft tracking method based on multi-source sensor data. By obtaining a set of to-be-predicted sensor data corresponding to a target aircraft at multiple moments, each of which includes multiple different sensors, and inputting the multiple sets of to-be-predicted sensor data into a trained trajectory tracking model, a flight trajectory tracking result corresponding to the target aircraft is obtained. The trajectory tracking model is trained according to a training set. Since the training set includes training sensor data sets corresponding to multiple moments and the true flight trajectory points corresponding to the training sensor data sets, the training sensor data sets can be spatially associated with the true flight trajectory points, enabling the trajectory tracking model to effectively track the flight trajectory of the target aircraft through sensor data, and also solving the problem in the prior art of relying on the ADS-B system to obtain prior information. And the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder. Since the bidirectional gated recurrent unit encoder can obtain rich time series feature information at multiple moments, the accuracy of tracking the flight trajectory of the target aircraft is improved.

[0054] In one embodiment, as Figure 2 shown, Figure 2 is a schematic flowchart of an aircraft tracking method based on multi-source sensor data provided by an embodiment of the present invention, which specifically includes the following steps:

[0055] S10: Obtain a set of to-be-predicted sensor data corresponding to a target aircraft at multiple moments.

[0056] Among them, the multiple moments refer to multiple moments included within a preset time period. The preset time period can be determined by setting the interval for collecting sensor data by the sensors during the flight of the target aircraft. And for each moment, multiple sensors included in the ground distributed sensor network will simultaneously collect sensor data during the flight of the target aircraft. That is, the set of to-be-predicted sensor data corresponding to each moment includes: to-be-predicted data of multiple different sensors.

[0057] The above to-be-predicted data includes: the position parameter, identification parameter, time parameter, and signal strength parameter of the sensor. The position parameter includes: the first longitude parameter, the first latitude parameter, and the earth height parameter of the sensor. However, it is not limited thereto. The present invention does not specifically limit, and those skilled in the art can set according to actual situations.

[0058] It should be noted that for the sensor data collected by multiple sensors included in the ground distributed sensor network at the same moment, a preset number of sensor data are randomly selected as the data to be predicted, for example, 3. However, this is not limited thereto, and the present invention does not specifically limit it. Those skilled in the art can set it according to the actual situation.

[0059] Exemplarily, the preset duration can be, for example, the duration of a flight of an aircraft, such as 1 hour. If the sensor data acquisition interval of the target aircraft during flight is set to 1 second, then there are 3600 moments. The sensor data of the aircraft are collected by 10 sensors included in the ground distributed sensor network, and 3 are randomly selected from the 10 sensor data as the data to be predicted. However, this is not limited thereto, and the present invention does not specifically limit it. Those skilled in the art can set it according to the actual situation.

[0060] S11: Input multiple datasets of sensors to be predicted into the trained trajectory tracking model to obtain the flight trajectory tracking result corresponding to the target aircraft.

[0061] Among them, the trajectory tracking model is trained according to a training set. The training set includes: training sensor datasets corresponding to multiple moments, and the true flight trajectory points corresponding to the training sensor datasets. The training sensor datasets include: training sensor data collected by multiple different sensors.

[0062] The above true flight trajectory points are determined according to the historical flight data of the aircraft. Refer to Figure 3 As shown, the trajectory tracking model includes: a feature embedding layer 11, a bidirectional gated recurrent unit encoder 12, and a gated recurrent unit decoder 13.

[0063] Among them, the feature embedding layer can be an Embedding layer. Through the Embedding layer, the input sensor data to be predicted are mapped into a high-dimensional feature space to realize the standardization processing of the sensor data to be predicted, so as to obtain high-dimensional feature vectors.

[0064] The bidirectional gated recurrent unit encoder 12 is used to obtain the spatio-temporal correlation between the sensor data and the historical true flight trajectory data of the aircraft, that is, the flight coordinate position, and can solve problems such as heterogeneity, non-linearity, clock drift, noise interference, and inaccurate position existing in the sensor data, so as to accurately track the flight trajectory of the target aircraft.

[0065] The gated recurrent unit decoder 13 is used to predict and track the flight trajectory corresponding to the target aircraft to obtain the flight trajectory tracking result corresponding to the target aircraft.

[0066] Optionally, on the basis of the above embodiments, in some embodiments of the present invention, for S11, it can be based on the formula Modeling is performed to obtain the flight trajectory tracking results corresponding to the target aircraft. Among them, S is a plurality of sensor data sets to be predicted, E Sensor (·) represents the function corresponding to the encoding process of the bidirectional gated recurrent unit encoder, H represents the output of the bidirectional gated recurrent unit encoder, and D Trajectory (·) represents a function corresponding to the decoding process of the gated recurrent unit decoder.

[0067] Optionally, based on the above embodiment, in some embodiments of the present invention, before executing S11, the following steps are further included:

[0068] S111: Construct a training set based on historical flight data of multiple aircraft and sensor data collected by multiple sensors.

[0069] The historical flight data refers to the real flight trajectory information of multiple aircraft, that is, the flight coordinate position. The historical flight data can be the aircraft message received by the ADS-B system ground station to determine the real flight coordinate position of the aircraft's historical flight in three-dimensional space. The flight coordinate position includes: the second longitude parameter, the second latitude parameter, and the earth's altitude parameter of the aircraft. For example, the aircraft coordinates at the tth moment can be expressed as: P t =(λ t ,φ t ,h t ), where, where, λ t Represents the second longitude parameter of the aircraft at time t; φ t represents the second latitude parameter of the aircraft at time t; h t represents the earth height parameter of the aircraft at the tth moment. For T moments, the aircraft coordinate position sequence of the aircraft can be expressed as:

[0070] The sensor data includes: location parameters, identification parameters, time parameters, and signal strength parameters of the sensor, and the location parameters include: first longitude parameters, first latitude parameters, and earth height parameters of the sensor. Exemplarily, the sensor data may be represented as: Wherein, ID represents the identification parameter of the sensor; The time parameter of the sensor is the tth moment, which is used to characterize the time when the signal reaches the sensor; It represents the signal strength parameter of the sensor at the tth moment, reflects the signal attenuation characteristics, and implies the relative distance information between the sensor and the aircraft; represents the position parameter of the sensor at time t, represents the first longitude parameter of the sensor at time t, Represents the first-dimensional parameter of the sensor at the t-th moment. Represents the altitude parameter of the sensor at the t-th moment. Further, map the sensor data to obtain the sensor data for training, that is Then for T moments, the sensor data feature sequence of the sensor data can be expressed as: However, it is not limited to this. The present invention is not specifically limited, and those skilled in the art can set it according to the actual situation.

[0071] It should be noted that the alignment between the historical flight data and the sensor data collected by multiple sensors can be achieved through the time parameter.

[0072] Specifically, obtain the historical flight data of the aircraft through the aircraft messages received by the ADS-B system, and according to the sensor data collected by multiple sensors. Based on the historical flight data of the aircraft and the sensor data collected by multiple sensors, construct a training set for training the trajectory tracking model.

[0073] Exemplarily, continuing with the above embodiment, for T moments, the aircraft coordinate position sequence: Sensor data feature sequence: Then the obtained training set is: Among them, A represents the total number of multiple aircraft.

[0074] Optionally, based on the above embodiment, in some embodiments of the present invention, one implementation manner of S111 can be:

[0075] S1111: For multiple aircraft, collect each sensor data set of each aircraft at multiple moments.

[0076] Among them, each sensor data set corresponding to each moment includes: sensor data collected by multiple different sensors. Exemplarily, continuing with the above embodiment, among the 10 sensors included in the ground distributed sensor network, randomly determine the sensor data collected by 3 sensors to construct a training set. However, it is not limited to this. The present invention is not specifically limited, and those skilled in the art can set it according to the actual situation.

[0077] S1112: According to the historical flight data, determine each flight coordinate position of each aircraft at multiple moments.

[0078] S1113: Based on each sensor data set corresponding to multiple aircraft at multiple moments respectively, and each flight coordinate position corresponding to each sensor data set, construct a training set.

[0079] Specifically, for multiple aircraft, each aircraft's corresponding sensor dataset at multiple moments is collected through multiple sensors, and based on historical flight data, each aircraft's flight coordinate position at multiple moments is determined. In this way, a training set for training a trajectory tracking model can be constructed through each aircraft's corresponding sensor dataset at multiple moments, as well as each flight coordinate position corresponding to each sensor dataset.

[0080] Optionally, based on the above embodiments, in some embodiments of the present invention, each sensor dataset corresponding to each aircraft at multiple moments is used as the training sensor dataset in the training set, and each flight coordinate position corresponding to each sensor dataset is used as the true flight trajectory point, that is, the training sensor data and the true flight trajectory points are in one-to-one correspondence.

[0081] Optionally, based on the above embodiments, in some embodiments of the present invention, in order to eliminate the differences in dimension and numerical range between different sensor data and flight coordinate positions, before performing S1113, it further includes:

[0082] Performing normalized data processing on each sensor dataset corresponding to each aircraft at multiple moments, as well as each flight coordinate position corresponding to each sensor dataset.

[0083] Specifically, the normalized data processing for each sensor dataset and each flight coordinate position corresponding to each sensor dataset can be achieved through a preset normalization formula, and the normalization formula can be defined by the following expression:

[0084]

[0085] where lon represents the longitude parameter, lat represents the latitude parameter; alt represents the altitude parameter of the earth's height.

[0086] It should be noted that when using the trained trajectory tracking model to obtain the flight trajectory corresponding to the target aircraft, the obtained flight trajectory tracking result is subjected to inverse normalization processing through the inverse normalization formula, so as to obtain the aircraft trajectory prediction result in the three-dimensional earth coordinates. The inverse normalization formula can be defined by the following expression:

[0087] lat = lat norm ·(lat max -lat min ) + lat min

[0088] It should be noted that the inverse normalization processing for lon and alt is the same, and will not be elaborated here.

[0089] S112: Input the training set into the initial trajectory tracking model, train the initial trajectory tracking model according to the teacher forcing strategy, and adjust the model parameters through the loss function until the model converges, obtaining the trained trajectory tracking model.

[0090] Among them, the teacher forcing strategy is a strategy for training recurrent neural networks or Seq2Seq models, mainly used to accelerate model convergence and stabilize model training. Its core idea is that during the model training process, the input of the decoder at each time step is not necessarily the result predicted in the previous time step, but may also directly use the true label as the input. By introducing the teacher forcing strategy, the convergence speed of the training model can be accelerated to a certain extent.

[0091] Specifically, input the constructed training set into the initial trajectory tracking model, train the initial trajectory tracking model according to the teacher forcing strategy, and adjust the model parameters through the loss function until the trajectory tracking model converges, ending the training, and obtaining the trained trajectory tracking model.

[0092] Optionally, based on the above embodiments, in some embodiments of the present invention, one implementation manner of S112 may be:

[0093] S1121: Input the training sensor data sets at N moments into the feature embedding layer, and extract the high-dimensional feature vectors corresponding to the training sensor data sets at each moment.

[0094] Specifically, input the training sensor data sets at N moments into the feature embedding layer, and use the feature embedding layer to perform high-dimensional mapping processing on the training sensor data to obtain the high-dimensional feature vectors corresponding to the training sensor data sets at each moment.

[0095] S1122: Input the N high-dimensional feature vectors into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the hidden state features.

[0096] Among them, the hidden state features include N time series features related to the high-dimensional feature vectors at N moments. Through the hidden state features, the feature information of the training sensor data sets at N moments can be fully obtained, which is beneficial to extracting more complete and rich feature information in the time series, thereby improving the accuracy of obtaining the flight trajectory tracking result of the target aircraft.

[0097] Optionally, based on the above embodiments, in some embodiments of the present invention, one implementation manner of S1122 may be:

[0098] S20: Input the first high-dimensional feature vector at the M-th moment, the first forward time series feature at the (M - 1)-th moment, and the second forward time series feature at the (M + 1)-th moment into a bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the time series feature at the M-th moment.

[0099] Where M = <N, exemplarily, N is 100, M is, for example, 20, but not limited thereto. The present invention does not specifically limit it, and those skilled in the art can set it according to the actual situation.

[0100] Specifically, input the first high-dimensional feature vector at the M-th moment, the first forward time series feature at the (M - 1)-th moment, and the second forward time series feature at the (M + 1)-th moment at N moments into a bidirectional gated recurrent unit encoder, and perform bidirectional encoding processing through the bidirectional gated recurrent unit encoder to obtain the time series feature at the M-th moment.

[0101] Optionally, based on the above embodiments, in some embodiments of the present invention, the bidirectional encoding processing includes forward encoding processing and backward encoding processing. Based on this, one implementation manner of S20 can be:

[0102] S201: Input the first high-dimensional feature vector and the first forward time series feature into a bidirectional gated recurrent unit encoder for forward encoding processing to obtain the forward time series feature at the M-th moment.

[0103] Specifically, input the first high-dimensional feature vector at the M-th moment and the first forward time series feature at the (M - 1)-th moment into a bidirectional gated recurrent unit encoder, and perform forward encoding processing through the bidirectional gated recurrent unit encoder to obtain the forward time series feature at the M-th moment.

[0104] Optionally, based on the above embodiments, in some embodiments of the present invention, the forward time series feature at the M-th moment can be obtained according to the formula where Input M represents the first high-dimensional feature vector at the M-th moment, represents the first forward time series feature at the (M - 1)-th moment.

[0105] S202: Input the first high-dimensional feature vector and the second forward time series feature into a bidirectional gated recurrent unit encoder for backward encoding processing to obtain the backward time series feature at the M-th moment.

[0106] Specifically, input the first high-dimensional feature vector at the M-th moment and the second forward time series feature at the (M + 1)-th moment into a bidirectional gated recurrent unit encoder, and perform backward encoding processing through the bidirectional gated recurrent unit encoder to obtain the backward time series feature at the M-th moment.

[0107] Optionally, based on the above embodiments, in some embodiments of the present invention, the reverse time series feature at the M-th moment can be obtained according to the formula where Input M represents the first high-dimensional feature vector at the M-th moment, represents the second forward time series feature at the (M + 1)-th moment.

[0108] It should be noted that the implementation processes for obtaining the forward time series features and reverse time series features at each of the N moments are the same, and will not be elaborated here.

[0109] S203: Perform splicing processing based on the forward time series feature and the reverse time series feature to obtain the time series feature at the M-th moment.

[0110] Specifically, splice the obtained forward time series feature at the M-th moment and the reverse time series feature at the M-th moment to obtain the time series feature at the M-th moment.

[0111] Optionally, based on the above embodiments, in some embodiments of the present invention, the time series feature at the M-th moment can be obtained according to the formula to obtain the time series feature at the M-th moment.

[0112] S21: Determine the hidden state feature according to the time series features at N moments.

[0113] Specifically, accumulate all the time series features at the obtained N moments to determine the hidden state feature. This hidden state feature is used as the hidden state of the initialized gated recurrent unit decoder.

[0114] S1123: After obtaining the hidden state feature, according to the initial flight trajectory point, the hidden state feature, and the true flight trajectory point, adopt the teacher forcing strategy and use the gated recurrent unit decoder to obtain the predicted flight trajectory points of the target aircraft at each moment.

[0115] Optionally, based on the above embodiments, in some embodiments of the present invention, based on this, one implementation manner of S1123 can be:

[0116] S30: When M is 0, input the initial flight trajectory point and the hidden state feature into the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment.

[0117] Among them, the initial flight trajectory point refers to the coordinate position of the aircraft set in advance, and this coordinate position is (0, 0, 0).

[0118] Specifically, at the moment when M is 0, that is, at the initial moment, the initial flight trajectory point and the hidden state feature are input into the gated recurrent unit decoder, and the initial flight trajectory point and the hidden state feature are decoded by the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment.

[0119] S31: According to the teacher forcing strategy, determine the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point, and update the hidden state feature by using the target decoding input feature to obtain the updated hidden state feature.

[0120] Specifically, after obtaining the predicted flight trajectory point at the next moment, in order to improve the training efficiency of the model, the teacher forcing strategy is adopted to determine the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point, and update the hidden state feature by using the target decoding input feature to obtain the updated hidden state feature.

[0121] Optionally, on the basis of the above embodiments, in some embodiments of the present invention, the update of the hidden state feature can be realized by the formula

[0122] Optionally, on the basis of the above embodiments, in some embodiments of the present invention, one implementation manner of determining the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point by adopting the teacher forcing strategy can be:

[0123] By hitting the probability P TF ∈[0,1], determine the input of the gated recurrent unit decoder among the predicted flight trajectory point and the true flight trajectory point. The process of the teacher forcing strategy is as follows:

[0124]

[0125] where b M ∈{0,1} represents a binary random variable, p TF ∈[0,1] represents the teacher forcing probability. When b M has a probability of p TF taking 1, determine the true flight trajectory point as the input of the gated recurrent unit decoder. Otherwise, determine the predicted flight trajectory point as the input of the gated recurrent unit decoder.

[0126] S32: Input the target decoding input feature and the updated hidden state feature into the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment, and return to execute to determine the target decoding input feature among the predicted flight trajectory point and the true flight trajectory point according to the teacher forcing strategy until the predicted flight trajectory points at N moments are obtained. ​

[0127] Specifically, through the teacher forcing strategy, the target decoding input features and the updated hidden state features are determined among the predicted flight trajectory points and the true flight trajectory points and input into the gated recurrent unit decoder. Decoding processing is performed through the gated recurrent unit decoder to obtain the predicted flight trajectory point at the next moment. After obtaining the predicted flight trajectory point at the next moment, continue to return and execute to determine the target decoding input features among the predicted flight trajectory points and the true flight trajectory points according to the teacher forcing strategy until all the predicted flight trajectory points at N moments are obtained.

[0128] S1124: Compare the predicted flight trajectory points with the true flight trajectory points, and adjust the model parameters according to the loss function until the model converges to obtain a trained trajectory tracking model.

[0129] Specifically, after obtaining the predicted flight trajectory points, compare the predicted flight trajectory points with the true flight trajectory points, and adjust the parameters of the model according to the loss function until the trajectory tracking model converges, end the training, and obtain a trained trajectory tracking model.

[0130] In this way, the aircraft tracking method based on multi-source sensor data provided in this embodiment obtains a to-be-predicted sensor data set including multiple different sensors corresponding to the target aircraft at multiple moments, inputs the multiple to-be-predicted sensor data sets into the trained trajectory tracking model, and obtains the flight trajectory tracking result corresponding to the target aircraft. The trajectory tracking model is obtained by training according to the training set. Since the training set includes the training sensor data sets corresponding to multiple moments and the true flight trajectory points corresponding to the training sensor data sets, the training sensor data sets can be spatially associated with the true flight trajectory points, so that the trajectory tracking model can effectively track the flight trajectory of the target aircraft through the sensor data, and also solves the problem of relying on the ADS-B system to obtain prior information in the prior art. And the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder. Since the bidirectional gated recurrent unit encoder can obtain rich time series feature information at multiple moments, the accuracy of tracking the flight trajectory of the target aircraft is improved.

[0131] It should be understood that although Figures 1 to 3 the steps in the flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, Figures 1 to 3At least some of the steps may include multiple sub-steps or multiple stages, and these sub-steps or stages do not necessarily need to be executed and completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages does not necessarily need to be sequential, but can be executed alternately or in turn with at least some of the sub-steps or stages of other steps or other steps.

[0132] In one embodiment, as Figure 4 shown, a device for tracking an aircraft based on multi-source sensor data is provided, including: an acquisition module 10 and a flight trajectory tracking result acquisition module 11.

[0133] Among them, the acquisition module 10 is configured to acquire a set of sensor data to be predicted corresponding to a target aircraft at multiple moments, where the set of sensor data to be predicted includes data to be predicted from multiple different sensors;

[0134] The flight trajectory tracking result acquisition module 11 is configured to input multiple sets of sensor data to be predicted into a trained trajectory tracking model to obtain a flight trajectory tracking result corresponding to the target aircraft;

[0135] Among them, the trajectory tracking model is obtained by training based on a training set. The training set includes: a set of training sensor data corresponding to multiple moments, and the true flight trajectory points corresponding to the set of training sensor data. The set of training sensor data includes: training sensor data collected by multiple different sensors. The trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder.

[0136] In the above embodiment, the acquisition module acquires a set of sensor data to be predicted corresponding to a target aircraft at multiple moments. The flight trajectory tracking result acquisition module inputs multiple sets of sensor data to be predicted into a trained trajectory tracking model to obtain a flight trajectory tracking result corresponding to the target aircraft. The trajectory tracking model is obtained by training based on a training set. Since the training set includes a set of training sensor data corresponding to multiple moments and the true flight trajectory points corresponding to the set of training sensor data, the training sensor data can be spatially associated with the true flight trajectory points, enabling the trajectory tracking model to effectively track the flight trajectory of the target aircraft through sensor data, and also solving the problem of relying on the ADS-B system to obtain prior information in the prior art. And the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder. Since the bidirectional gated recurrent unit encoder can obtain rich time series feature information at multiple moments, the accuracy of tracking the flight trajectory of the target aircraft is improved.

[0137] For the specific limitations of the aircraft tracking device based on multi-source sensor data, reference can be made to the limitations of the aircraft tracking method based on multi-source sensor data in the foregoing text, which will not be elaborated herein. Each module in the above server can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0138] An embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it can implement the aircraft tracking method based on multi-source sensor data provided by the embodiment of the present invention. For example, when the processor executes the computer program, it can implement Figures 1 to 3 the technical solutions of any of the illustrated method embodiments. The implementation principles and technical effects are similar and will not be elaborated herein.

[0139] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, a database, or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static random access memory (SRAM) and dynamic random access memory (DRAM).

[0140] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0141] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. An aircraft tracking method based on multi-source sensor data, characterized in that, Including: Obtain the to-be-predicted sensor data sets respectively corresponding to a target aircraft at multiple moments, where the to-be-predicted sensor data sets include to-be-predicted data of multiple different sensors; Input the multiple to-be-predicted sensor data sets into a trained trajectory tracking model to obtain a flight trajectory tracking result corresponding to the target aircraft; Among them, the trajectory tracking model is obtained by training based on a training set, and the training set includes: training sensor data sets respectively corresponding to multiple moments, and real flight trajectory points corresponding to the training sensor data sets. The training sensor data sets include: training sensor data collected by multiple different sensors. The trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder.

2. The method according to claim 1, wherein Before the step of inputting the multiple to-be-predicted sensor data sets into a trained trajectory tracking model to obtain a flight trajectory tracking result corresponding to the target aircraft, it further includes: Construct a training set according to the historical flight data of multiple aircraft and the sensor data collected by multiple sensors; Input the training set into an initial trajectory tracking model, train the initial trajectory tracking model according to the teacher forcing strategy, and adjust the model parameters through a loss function until the model converges to obtain the trained trajectory tracking model.

3. The method according to claim 2, wherein The step of constructing a training set according to the historical flight data of multiple aircraft and the sensor data collected by multiple sensors includes: For multiple aircraft, collect each sensor data set of each aircraft at multiple moments, where each sensor data set corresponding to each moment includes: sensor data collected by multiple different sensors; Determine the flight coordinate position of each aircraft at multiple moments according to the historical flight data; Construct a training set according to each sensor data set respectively corresponding to multiple aircraft at multiple moments and each flight coordinate position corresponding to each sensor data set.

4. The method according to claim 3, characterized in that, Before the step of constructing a training set according to each sensor data set respectively corresponding to multiple aircraft at multiple moments and each flight coordinate position corresponding to each sensor data set, it further includes: Perform normalized data processing on each sensor data set respectively corresponding to multiple aircraft at multiple moments and each flight coordinate position corresponding to each sensor data set.

5. The method according to claim 4, wherein The step of inputting the training set into an initial trajectory tracking model, training the initial trajectory tracking model according to the teacher forcing strategy, and adjusting the model parameters through a loss function until the model converges to obtain the trained trajectory tracking model includes: Input the training sensor data sets at N moments into the feature embedding layer to extract high-dimensional feature vectors corresponding to the training sensor data sets at each moment; Input the N high-dimensional feature vectors into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain hidden state features, where the hidden state features include N time series features related to the high-dimensional feature vectors at N moments; After obtaining the hidden state features, according to the initial flight trajectory points, the hidden state features, and the true flight trajectory points, adopting the teacher forcing strategy, through the gated recurrent unit decoder, obtain the predicted flight trajectory points of the target aircraft at each moment; Compare the predicted flight trajectory points with the true flight trajectory points, and adjust the model parameters according to the loss function until the model converges, and obtain the trained trajectory tracking model.

6. The method according to claim 5, wherein The inputting the N high-dimensional feature vectors into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain hidden state features includes: Input the first high-dimensional feature vector at the Mth moment, the first forward time series feature at the (M - 1)th moment, and the second forward time series feature at the (M + 1)th moment into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the time series feature at the Mth moment, where M < N; Determine the hidden state features according to the time series features at N moments.

7. The method according to claim 6, wherein The bidirectional encoding processing includes forward encoding processing and backward encoding processing. The inputting the first high-dimensional feature vector at the Mth moment, the first forward time series feature at the (M - 1)th moment, and the second forward time series feature at the (M + 1)th moment into the bidirectional gated recurrent unit encoder for bidirectional encoding processing to obtain the time series feature at the Mth moment includes: Input the first high-dimensional feature vector and the first forward time series feature into the bidirectional gated recurrent unit encoder for forward encoding processing to obtain the forward time series feature at the Mth moment; Input the first high-dimensional feature vector and the second forward time series feature into the bidirectional gated recurrent unit encoder for backward encoding processing to obtain the backward time series feature at the Mth moment; Perform splicing processing according to the forward time series feature and the backward time series feature to obtain the time series feature at the Mth moment.

8. The method according to claim 5, wherein The obtaining the predicted flight trajectory points of the target aircraft at each moment by adopting the teacher forcing strategy through the gated recurrent unit decoder according to the initial flight trajectory points, the hidden state features, and the true flight trajectory points after obtaining the hidden state features includes: At the moment when M is 0, input the initial flight trajectory points and the hidden state features into the gated recurrent unit decoder to obtain the predicted flight trajectory points at the next moment; According to the teacher forcing strategy, determine the target decoding input feature among the predicted flight trajectory points and the true flight trajectory points, and update the hidden state features by using the target decoding input feature to obtain updated hidden state features; Input the target decoding input feature and the updated hidden state features into the gated recurrent unit decoder to obtain the predicted flight trajectory points at the next moment, and return to execute determining the target decoding input feature among the predicted flight trajectory points and the true flight trajectory points according to the teacher forcing strategy until the predicted flight trajectory points at N moments are obtained.

9. An aircraft tracking device based on multi-source sensor data, characterized in that, including: An acquisition module, configured to acquire a to-be-predicted sensor data set corresponding to a target aircraft at multiple moments, wherein the to-be-predicted sensor data set includes to-be-predicted data of multiple different sensors; A flight trajectory tracking result acquisition module, configured to input the multiple to-be-predicted sensor data sets into a trained trajectory tracking model to acquire a flight trajectory tracking result corresponding to the target aircraft; Wherein, the trajectory tracking model is obtained by training according to a training set, the training set includes: training sensor data sets corresponding to multiple moments, and real flight trajectory points corresponding to the training sensor data sets, the training sensor data sets include: training sensor data collected by multiple different sensors, and the trajectory tracking model includes: a feature embedding layer, a bidirectional gated recurrent unit encoder, and a gated recurrent unit decoder.

10. An electronic device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the aircraft tracking method based on multi-source sensor data according to any one of claims 1 to 8 are implemented.

Images