Secondary Radar Track Processing Method Based on PCA-LSTM Algorithm

Abstract

The present application relates to a method and device for processing secondary radar tracks based on the PCA-LSTM algorithm. The method includes: obtaining historical secondary radar data and ADS-B position data, and associating the historical secondary radar data and the ADS-B position data. Training a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data by using principal component analysis. The target track prediction module is a long short-term memory neural network model based on time series, which is used to predict the future azimuth and distance of the target. Inputting the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track. It not only realizes the tracking of air targets of different maneuver types, but also improves the efficiency of secondary radar track prediction.

Description

Technical Field

[0001] The present application relates to the technical field of radar track processing, and in particular to a secondary radar track processing method based on the PCA-LSTM algorithm. Background Art

[0002] In the process of secondary radar track processing, the tracking of maneuvering targets has always been a key and difficult problem. The biggest difficulty lies in the uncertainty of target motion. Especially for air targets of different maneuver types, there is a high probability that the motion models do not match, which affects the tracking accuracy and stability of the targets.

[0003] In the prior art, mainly the constant velocity model (CV), constant acceleration model (CA), Singer model, jerk model, interacting multiple model (IMM), etc. are adopted. Among them, the CV and CA models use uniform and constant acceleration linear motion modeling, only considering the random interference noise of environmental factors. These two models are simple and practical, and have good tracking effects for general non-maneuvering or weakly maneuvering targets. The Singer model breaks through the shackles of modeling the maneuver term with Gaussian white noise and realizes the modeling process of the target acceleration with a zero-mean random process with exponential autocorrelation. However, the Singer model does not "break out" of the assumption that the mean value of the target maneuver acceleration is zero and the acceleration probability density function follows a uniform distribution, and there is still a certain inaccuracy in the description of maneuvering targets. By analogy with the noise modeling process of the Singer model, the Jerk model extends the derivative of the target state vector to the third order and uses a zero-mean random process with exponential autocorrelation to model the acceleration change rate of the target. Compared with the Singer model, at the moment of large target maneuvers, the jerk model improves the tracking accuracy of the target, but it is still difficult to face air targets of different maneuver types. To further solve this problem, the interacting multiple model (IMM) algorithm came into being. It uses two or more motion models to match the states that the target may appear in maneuvering motion, and then realizes the estimation of the target state through weighted fusion of the filtering results of multiple models. This model usually combines non-maneuvering models and maneuvering models, and controls the interaction and conversion between models through a Markov chain, so as to achieve comprehensive adaptive tracking of the target. Its modular characteristics enable this model to perform more flexible combination designs. However, the computational complexity of this algorithm also increases with the increase in the number of models, limiting the scope of engineering applications.

[0004] Therefore, in the related art, there is an urgent need for a method that can track air targets of different maneuver types and improve the efficiency of secondary radar track prediction. Summary of the Invention

[0005] Based on this, it is necessary to provide a secondary radar track processing method based on the PCA-LSTM algorithm for the above technical problems, which can track air targets of different maneuver types and improve the efficiency of secondary radar track prediction.

[0006] In a first aspect, the present application provides a secondary radar track processing method based on the PCA-LSTM algorithm. The method includes:

[0007] Obtain historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and ADS-B position data;

[0008] Train a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term time memory neural network model based on time series, which is used to predict the future azimuth and distance of the target;

[0009] Input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track.

[0010] Optionally, in an embodiment of the present application, the obtaining of historical secondary radar data and ADS-B position data includes:

[0011] Obtain the historical secondary radar data and ADS-B position data of low-speed maneuvering targets, medium-speed maneuvering targets, and high-speed maneuvering targets based on flight speed and acceleration respectively;

[0012] Obtain the historical secondary radar data and ADS-B position data of nearby targets, medium-nearby targets, medium-far targets, and far targets based on distance respectively;

[0013] Obtain the historical secondary radar data and ADS-B position data of targets near the left beam edge, near the beam center, and near the right beam edge based on azimuth respectively.

[0014] Optionally, in an embodiment of the present application, the associating of the historical secondary radar data and ADS-B position data includes:

[0015] Associate the historical secondary radar data and ADS-B data of the same target at the same moment according to the target code or S-mode address and time.

[0016] Optionally, in an embodiment of the present application, the dimensionality reduction module includes:

[0017] Construct a sample matrix;

[0018] Perform centering processing on the sample matrix and calculate the covariance matrix of the sample matrix;

[0019] Perform eigenvalue decomposition based on the covariance matrix to determine the required dimensionality reduction dimension;

[0020] Determine the feature data after dimensionality reduction based on the dimensionality reduction dimension.

[0021] Optionally, in an embodiment of the present application, the training of the target track prediction module includes:

[0022] Input the feature data after dimensionality reduction into the target track prediction module to obtain predicted ADS-B position data;

[0023] Determine the difference based on the predicted ADS-B position data and the corresponding associated ADS-B position data;

[0024] Modify the neuron weights of the target track prediction module based on the difference.

[0025] Optionally, in an embodiment of the present application, the training of the target track prediction module further includes:

[0026] Divide the feature data after dimensionality reduction and the corresponding associated ADS-B data into a training set and a test set;

[0027] Modify the neuron weights of the target track prediction module based on the effect obtained from the training set, and evaluate the modified secondary radar track prediction model based on the test set.

[0028] Optionally, in an embodiment of the present application, the evaluation of the modified secondary radar track prediction model based on the test set includes:

[0029] The model evaluation metrics are root mean square error, correlation coefficient, mean absolute error, and mean bias error. Among them, the calculation formula for root mean square error is:

[0030]

[0031] The calculation formula for the correlation coefficient is:

[0032]

[0033] The calculation formula for mean absolute error is:

[0034]

[0035] The calculation formula for mean bias error is:

[0036]

[0037] Among them, y i is the azimuth or distance data (true value) of the maneuvering target's ADS - B, and y pred is the azimuth or distance data (predicted value) of the target predicted by the PCA - LSTM model. is the average of the target true value, m is the total number of samples, and i represents the sample serial number.

[0038] In a second aspect, the present application also provides a secondary radar track processing device based on the PCA - LSTM algorithm.

[0039] The device includes:

[0040] A data acquisition module, configured to acquire historical secondary radar data and ADS - B position data, and associate the historical secondary radar data and ADS - B position data;

[0041] A secondary radar track prediction model training module, configured to train a secondary radar track prediction model based on the associated historical secondary radar data and ADS - B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, configured to determine feature data based on the historical secondary radar data, and perform dimensionality reduction processing on the feature data by using principal component analysis. The target track prediction module is a long - short - term time - memory neural network model based on time series, configured to predict the future azimuth and distance of the target;

[0042] A target track determination module, configured to input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track.

[0043] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and the processor executes the steps of the methods described in the above respective embodiments.

[0044] In a fourth aspect, the present application also provides a computer - readable storage medium. The computer - readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the methods described in the above respective embodiments.

[0045] The above secondary radar track processing method, device, computer equipment and storage medium based on the PCA-LSTM algorithm obtain historical secondary radar data and ADS-B position data, and correlate the historical secondary radar data and ADS-B position data. Then, based on the correlated historical secondary radar data and ADS-B position data, a secondary radar track prediction model is trained. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term time memory neural network model based on time series, which is used to predict the future azimuth and distance of the target. Finally, the secondary radar data of the target to be predicted is input into the trained secondary radar track prediction model to obtain the target track. That is to say, in the process of secondary radar track processing, by combining the principal component analysis method to perform dimensionality reduction processing on the input feature data, and using an end-to-end long short-term memory neural network model (LSTM) based on time series, by taking the original decoded data and plot data of various secondary radar maneuvering targets as the input of the training model and the ADS-B data with higher target accuracy as the output data of the model training, a large number of trainings are carried out on the established LSTM model, and the model parameters are continuously optimized, which not only realizes the tracking of air targets of different maneuvering types, but also improves the efficiency of secondary radar track prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 FIG. is an application environment diagram of the secondary radar track processing method based on the PCA-LSTM algorithm in an embodiment;

[0047] Figure 2 FIG. is a schematic flowchart of the secondary radar track processing method based on the PCA-LSTM algorithm in an embodiment;

[0048] Figure 3 FIG. is a schematic diagram of the covariance eigenvalue of sample data during principal component analysis in an embodiment;

[0049] Figure 4 FIG. is a schematic diagram of the internal structure and data flow of a single neuron in LSTM in an embodiment;

[0050] Figure 5 FIG. is a schematic diagram of the prediction effect of the training set and test set of the PCA-LSTM model in an embodiment;

[0051] Figure 6 FIG. is a schematic diagram of the performance of the PCA-LSTM model under four evaluation indexes in an embodiment;

[0052] Figure 7It is a schematic flowchart of the specific steps of the secondary radar track processing method based on the PCA-LSTM algorithm in an embodiment;

[0053] Figure 8 It is a structural block diagram of the secondary radar track processing device based on the PCA-LSTM algorithm in an embodiment;

[0054] Figure 9 It is the internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] The secondary radar track processing method based on the PCA-LSTM algorithm provided in the embodiments of the present application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0057] In one embodiment, as Figure 2 shown, a secondary radar track processing method based on the PCA-LSTM algorithm is provided. Taking the method applied to the Figure 1 server as an example, the method includes the following steps:

[0058] S201: Obtain historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and ADS-B position data.

[0059] In the embodiments of the present application, first, historical secondary radar data and ADS-B position data are obtained. The historical secondary radar data refers to the original decoded data of targets with different maneuver types and targets with multiple maneuver types. The information carried by the original decoded data includes: sum-channel amplitude, difference-channel amplitude, control-channel amplitude, sign bit, wave position number, time, target code, and target distance. The information carried by the ADS-B position data includes: identity code, target code, time, target distance, azimuth, and altitude. Then, the historical secondary radar data and the ADS-B position data are associated, that is, the original decoded data of each target and the corresponding ADS-B position data are associated one by one to form sample data.

[0060] S203: Train a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term time memory neural network model based on time series, which is used to predict the future azimuth and distance of the target.

[0061] In the embodiments of the present application, for the sample data obtained after association, that is, the historical secondary radar data and the ADS-B position data, the historical secondary radar data is used as the input of the secondary radar track prediction model, and the ADS-B position data is used as the true reference data of the output data to train the dimensionality reduction module and the target track prediction module of the secondary radar track prediction model. Among them, the dimensionality reduction module is mainly composed of a PCA algorithm model. For the same target, the feature data related to the track includes: sum-channel amplitude, difference-channel amplitude, sign bit, wave position number, time, response times, target distance, and azimuth. Among them, the azimuth information needs to be obtained through secondary calculation of the sum-channel amplitude, difference-channel amplitude, sign bit, and querying the OBA table. The dimensionality reduction module mainly performs dimensionality reduction processing on the feature data using principal component analysis (PCA), and determines the required dimensionality reduction dimension according to the eigenvalue size of the sample covariance matrix to ensure that the feature data after dimensionality reduction processing can still accurately reflect the true sample feature data. The target track prediction module is mainly composed of a long short-term time memory neural network model based on time series. The historical secondary radar data after dimensionality reduction processing is used as the input of the target track prediction module, and the ADS-B data with high position accuracy is used as the true reference data of the output data of the target aircraft prediction module. By continuously adjusting the parameters of the model according to the difference between the predicted value and the true value of the target position obtained after each training, the optimal weight of the model is finally obtained through continuous iteration and optimization.

[0062] S205: Input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track.

[0063] In the embodiment of the present application, the secondary radar data of the target to be predicted, that is, the original decoded data, is input into the trained secondary radar track prediction model, and the target track is output, that is, the target azimuth and distance after Δt are predicted backward at the current time t.

[0064] In the above secondary radar track processing method based on the PCA-LSTM algorithm, first, historical secondary radar data and ADS-B position data are obtained, and the historical secondary radar data and ADS-B position data are associated. Then, based on the associated historical secondary radar data and ADS-B position data, a secondary radar track prediction model is trained. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term time memory neural network model based on time series, which is used to predict the future azimuth and distance of the target. Finally, the secondary radar data of the target to be predicted is input into the trained secondary radar track prediction model to obtain the target track. That is to say, in the process of secondary radar track processing, by combining the principal component analysis method to perform dimensionality reduction processing on the input feature data, and using an end-to-end long short-term memory neural network model (LSTM) based on time series. By using the original decoded data and plot data of various secondary radar maneuvering targets as the input of the training model, and using the ADS-B data with higher target accuracy as the output data of model training, a large amount of training is carried out on the established LSTM model, and the model parameters are continuously optimized, which not only realizes tracking air targets of different maneuvering types, but also improves the secondary radar track prediction efficiency.

[0065] In an embodiment of the present application, the obtaining of historical secondary radar data and ADS-B position data includes:

[0066] S301: Based on the flight speed and acceleration, obtain the historical secondary radar data and ADS-B position data of low-speed maneuvering targets, medium-speed maneuvering targets, and high-speed maneuvering targets respectively.

[0067] S303: Based on the distance, obtain the historical secondary radar data and ADS-B position data of nearby targets, medium-nearby targets, medium-far targets, and far targets respectively.

[0068] S305: Based on the azimuth, obtain the historical secondary radar data and ADS-B position data of targets near the left beam edge, near the beam center, and near the right beam edge respectively.

[0069] In one embodiment of the present application, to ensure the diversity of training data, a large amount of track data of different maneuver type targets and multi-maneuver type targets is selected, including: maneuver targets are classified into low-speed maneuver targets, medium-speed maneuver targets, and high-speed maneuver targets according to flight speed and acceleration. For each type of maneuver target, in terms of distance: near targets within 0 - 50 kilometers, medium-near targets between 50 - 150 kilometers, medium-far targets between 150 - 250 kilometers, and far targets over 250 kilometers need to be selected respectively. In terms of azimuth, targets near the left beam edge, near the beam center, and near the right beam edge need to be selected respectively.

[0070] In this embodiment, by obtaining a large amount of track data of different maneuver type targets and multi-maneuver type targets, the diversity of training data is ensured.

[0071] In one embodiment of the present application, the associating the historical secondary radar data and ADS-B position data includes:

[0072] Associating the historical secondary radar data and ADS-B data of the same target at the same moment according to the target code or S-mode address and time.

[0073] In one embodiment of the present application, using the pre-written data analysis software, the secondary radar data of the same target and the corresponding ADS-B target data are screened according to the target code or S-mode address. After screening out the same target, the secondary radar data and ADS-B data of the target at the same moment are screened according to time, so that the two types of data of the target are in one-to-one correspondence in time. If there is a time difference, time compensation processing is required, and a new format data file is generated. Select different target codes or S-mode addresses, and repeat the above steps until the secondary radar data and ADS-B data of all targets in the collected data are associated, and the sample data D = {x1, x2, x3,... x m}, x i = {x1, x2, x3,... x d , y1, y2} is generated, where x i is the secondary radar data and y i is the ADS-B data.

[0074] In this embodiment, by associating the historical secondary radar data and ADS-B data of the same target at the same moment according to the target code or S-mode address and time, the secondary radar data and ADS-B data of the target can be associated one by one, which is convenient for subsequent training of the prediction model and improves the overall efficiency.

[0075] In one embodiment of the present application, the dimensionality reduction module includes:

[0076] S401: Construct a sample matrix.

[0077] S403: Centralize the sample matrix and calculate the covariance matrix of the sample matrix.

[0078] S405: Perform eigenvalue decomposition based on the covariance matrix to determine the required dimensionality reduction dimension.

[0079] S407: Determine the feature data after dimensionality reduction based on the dimensionality reduction dimension.

[0080] In an embodiment of the present application, eight feature items including the sum channel amplitude, difference channel amplitude, sign bit, wave position number, time, response times, target distance, and azimuth in the secondary radar data are used as the input data for model training. The PCA is used to perform dimensionality reduction processing on the feature data input to the model. The dimensionality reduction process includes: First, construct a sample matrix X = {x1, x2, x3,... x m}, the number of samples is m, and x i is a vector with a dimension of d. Then, centralize all samples X, and calculate the sample covariance matrix XX T . Then, perform eigenvalue decomposition on the covariance matrix XX T . The eigenvectors W = {w1, w2, w3,... w s′} corresponding to the largest d' (the actual dimension of the sample is d) eigenvalues. Finally, the feature data Z after dimensionality reduction can be obtained by linearly transforming the original sample X, Z = W T X. The sample data Z after dimensionality reduction will be used as the data input of the LSTM model. As Figure 3 shown, it is the eigenvalues of the sample data covariance during principal component analysis in an embodiment, showing the proportion of the eight eigenvalues after eigenvalue decomposition of the sample matrix covariance. It can be seen from the figure that the proportions of the first three eigenvalues are 73.53%, 19.34%, and 6.30% respectively, which means that 99.17% of the data information in the sample matrix X has been retained in the directions of the eigenvectors corresponding to the first three eigenvalues. That is, the sample matrix X with the original eight feature dimensions can be reduced to three dimensions while retaining 99.17% of the data information, which will greatly reduce the computational amount of the LSTM model. In addition, the remaining information of less than 1% largely belongs to noise information. Appropriately discarding it is also beneficial to reducing the interference of noise information on the model and accelerating the convergence speed of the model.

[0081] In this embodiment, by using PCA to perform dimensionality reduction processing on the feature data input to the model, the operation time of the model is reduced, the efficiency is improved, and unnecessary noise is filtered out.

[0082] In one embodiment of the present application, the training of the target track prediction module includes:

[0083] S501: Input the dimension-reduced feature data into the target track prediction module to obtain predicted ADS-B position data.

[0084] S503: Determine the difference based on the predicted ADS-B position data and the corresponding associated ADS-B position data.

[0085] S505: Modify the neuron weights of the target track prediction module based on the difference.

[0086] In one embodiment of the present application, the eight feature items of the amplitude of the sum channel, the amplitude of the difference channel, the sign bit, the wave position number, the time, the number of responses, the target distance, and the azimuth in the dimension-reduced secondary radar data are used as the input data for model training. Input them into the target track prediction module to obtain predicted ADS-B position data, that is, predicted target distance and azimuth data. The real target distance and real azimuth data in the corresponding associated ADS-B position data are used as the real comparison data of the model training output data. Determine the difference between the predicted ADS-B position data and the corresponding associated ADS-B position data, that is, the difference between the predicted target distance and the real target distance, and the difference between the predicted azimuth data and the real azimuth data. Then modify the neuron weights of the target track prediction module based on the size of the difference.

[0087] In this embodiment, by inputting the dimension-reduced feature data into the target track prediction module to obtain predicted ADS-B position data, determining the difference based on the predicted ADS-B position data and the corresponding associated ADS-B position data, and modifying the neuron weights of the target track prediction module based on the difference, and using the real ADS-B position data as the comparison data to train the model, the training effect of the model can be improved, and accurate tracking and prediction of maneuvering targets can be achieved.

[0088] In one embodiment of the present application, the training of the target track prediction module further includes:

[0089] S601: Divide the dimension-reduced feature data and the corresponding associated ADS-B data into a training set and a test set.

[0090] S603: Modify the neuron weights of the target track prediction module based on the effect obtained from the training set, and evaluate the modified secondary radar track prediction model based on the test set.

[0091] In one embodiment of the present application, 80% of the dimension-reduced feature data and the corresponding associated ADS-B data are divided into a training set, and 20% are divided into a test set. All the data in the training set and the test set are normalized, and the range of the normalized data is [0, 1]. As Figure 4 shown, it is a schematic diagram of the internal structure and data flow of a single neuron of LSTM. The LSTM model is set to 5 layers, namely the input layer, the LSTM layer, the ReLU layer, the fully connected fc layer, and the output layer. The number of nodes in the input layer is set to 3, the number of hidden units in the LSTM layer is 4, the number of hidden units in the ReLU activation layer is 4, the number of hidden units in the fully connected fc layer is 2, and the number of nodes in the output layer is 2.

[0092] Among them, the LSTM layer contains four neuron nodes, and each node contains a forget gate, an input gate, and an output gate. The calculation formula of the forget gate is:

[0093] f t = σ(W f [h t-1 , x t ) + b f )

[0094] Among them, σ is the sigmoid function, and its output is a value between 0 and 1. The use of this function can filter out unimportant input information and retain important input information.

[0095] The calculation formula of the input gate is:

[0096] i t = σ(W i [h t-1 , x t ) + b i )

[0097] C′ t = tanh(W C [h t-1 , x t ) + b C )

[0098] C t = f t C t-1 + i t C′ t

[0099] The calculation formula of the output gate is:

[0100] o t = σ(W o [h t-1 , x t ) + b o )

[0101] h t = o t tanh(C t )

[0102] After that, the parameters of the model are continuously optimized based on the effects obtained from the training set, and the effects of the modified secondary radar track prediction model are evaluated based on the test set. The model optimization method uses the stochastic gradient descent method, and the maximum number of iterations is 1500 times. To accelerate the model training speed and avoid model oscillation or divergence at the same time, the initial learning rate of the model is set to 0.01, and after 1200 times of training, the learning rate is reduced to 0.001.

[0103] In this embodiment, due to the special structural combination of the forgetting gate, input gate, and output gate of the LSTM model, it has both short-term and long-term memories, effectively avoiding the problem of gradient disappearance in the recurrent neural network. Therefore, in the process of processing secondary radar track data, the trained LSTM model can simultaneously remember the long-term and short-term states of maneuvering targets, so as to achieve accurate prediction of target tracks.

[0104] In an embodiment of the present application, the evaluating the modified secondary radar track prediction model based on the test set includes:

[0105] The model evaluation indexes are root mean square error, correlation coefficient, mean absolute error, and mean bias error. Among them, the calculation formula for the root mean square error is:

[0106]

[0107] The calculation formula for the correlation coefficient is:

[0108]

[0109] The calculation formula for the mean absolute error is:

[0110]

[0111] The calculation formula for the mean bias error is:

[0112]

[0113] Among them, y i is the ADS-B azimuth or distance data (true value) of the maneuvering target, y pred is the target azimuth or distance data (predicted value) predicted by the PCA-LSTM model, is the average of the target true values, m is the total number of samples, and i represents the sample serial number.

[0114] In one embodiment of the present application, for the evaluation of the secondary radar track prediction model, four indicators, namely root mean square error, correlation coefficient, mean absolute error, and mean bias error, are mainly used to evaluate the prediction effect of the model. As Figure 5 shown, it is a schematic diagram of the prediction effect of the PCA-LSTM model training set and test set. As can be seen from the figure, on the training set, the root mean square error RMSE of the model is 0.35043. The graph plotted with the true value as the x-axis and the predicted value as the y-axis is evenly distributed on the ray with an angle of 45 degrees to the x-axis, indicating that the predicted value and the true value are very close and the effect is good. Similarly, on the test set, the root mean square error RMSE of the model is 0.36306, which is very close to the training set. The graph plotted with the true value as the x-axis and the predicted value as the y-axis is also evenly distributed on the ray with an angle of 45 degrees to the x-axis, indicating that the model exhibits excellent generalization ability. As Figure 6 shown, it is a schematic diagram of the performance of the PCA-LSTM model under four evaluation indicators, which more intuitively shows the performance of the model under the other three evaluation indicators in addition to the RMSE indicator. Under the training set, R2 = 0.99827, MAE = 0.2866, MBE = 0.11107; under the test set, R2 = 0.99842, MAE = 0.28449, MBE = 0.10927; the above indicators all indicate that the model has excellent prediction effect, strong generalization ability, and very small deviation between the predicted value and the true value.

[0115] In this embodiment, the prediction effect of the model is evaluated through four indicators: root mean square error, correlation coefficient, mean absolute error, and mean bias error, which ensures the prediction ability of the model.

[0116] Next, a specific embodiment is used to illustrate the specific implementation steps of the secondary radar track processing method based on the PCA-LSTM algorithm of the present application. As Figure 7 shown, first, S701, obtain historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and ADS-B position data. Specifically, S703-S709, based on flight speed and acceleration, obtain historical secondary radar data and ADS-B position data of low-speed maneuvering targets, medium-speed maneuvering targets, and high-speed maneuvering targets respectively. Based on distance, obtain historical secondary radar data and ADS-B position data of nearby targets, medium-near targets, medium-far targets, and far targets respectively. Based on azimuth, obtain historical secondary radar data and ADS-B position data of targets near the left beam edge, near the beam center, and near the right beam edge respectively. According to the target code or S-mode address and time, associate the historical secondary radar data and ADS-B data of the same target at the same moment.

[0117] After that, in S711, a secondary radar track prediction model is trained based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term memory neural network model based on time series, which is used to predict the future azimuth and distance of the target.

[0118] Specifically, in S713 - S719, the dimensionality reduction module includes: constructing a sample matrix; performing centering processing on the sample matrix and calculating the covariance matrix of the sample matrix; performing eigenvalue decomposition based on the covariance matrix to determine the required dimensionality reduction dimension; and determining the dimensionality-reduced feature data based on the dimensionality reduction dimension. After that, in S721 - S725, the dimensionality-reduced feature data is input into the target track prediction module to obtain predicted ADS-B position data; the difference is determined based on the predicted ADS-B position data and the corresponding associated ADS-B position data; and the neuron weights of the target track prediction module are modified based on the difference.

[0119] In S727 - S729, the dimensionality-reduced feature data and the corresponding associated ADS-B data are divided into a training set and a test set; the neuron weights of the target track prediction module are modified based on the effect obtained from the training set, and the modified secondary radar track prediction model is evaluated based on the test set. Specifically, the model evaluation metrics are root mean square error, correlation coefficient, mean absolute error, and mean bias error. Among them, the calculation formula for root mean square error is:

[0120]

[0121] The calculation formula for the correlation coefficient is:

[0122]

[0123] The calculation formula for mean absolute error is:

[0124]

[0125] The calculation formula for mean bias error is:

[0126]

[0127] Among them, y i is the ADS-B azimuth or distance data (true value) of the maneuvering target, and y pred is the target azimuth or distance data (predicted value) predicted by the PCA-LSTM model. is the average of the target true values, m is the total number of samples, and i represents the sample serial number.

[0128] Finally, in S731, input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track.

[0129] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least some of the steps or stages in other steps or other steps.

[0130] Based on the same inventive concept, the embodiments of the present application also provide a secondary radar track processing device based on the PCA-LSTM algorithm for implementing the above-mentioned secondary radar track processing method based on the PCA-LSTM algorithm. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the secondary radar track processing device based on the PCA-LSTM algorithm provided below can refer to the limitations on the secondary radar track processing method based on the PCA-LSTM algorithm in the above text, and will not be repeated here.

[0131] In one embodiment, as Figure 8 shown, a secondary radar track processing device 800 based on the PCA-LSTM algorithm is provided, including: a data acquisition module 801, a secondary radar track prediction model training module 803, and a target track determination module 805, where:

[0132] The data acquisition module 801 is used to acquire historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and ADS-B position data.

[0133] The secondary radar track prediction model training module 803 is used to train a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term memory neural network model based on time series, which is used to predict the future azimuth and distance of the target.

[0134] The target track determination module 805 is used to input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track.

[0135] In an embodiment of the present application, the data acquisition module is further used for:

[0136] Based on the flight speed and acceleration, respectively obtain the historical secondary radar data and ADS-B position data of low-speed maneuvering targets, medium-speed maneuvering targets, and high-speed maneuvering targets;

[0137] Based on the distance, respectively obtain the historical secondary radar data and ADS-B position data of nearby targets, medium-nearby targets, medium-far targets, and far targets;

[0138] Based on the azimuth, respectively obtain the historical secondary radar data and ADS-B position data of targets near the left beam edge, near the beam center, and near the right beam edge.

[0139] In an embodiment of the present application, the data acquisition module is further used for:

[0140] Associate the historical secondary radar data and ADS-B data of the same target at the same moment according to the target code or S-mode address and time.

[0141] In an embodiment of the present application, the secondary radar track prediction model training module is further used for:

[0142] Construct a sample matrix;

[0143] Perform centering processing on the sample matrix and calculate the covariance matrix of the sample matrix;

[0144] Based on the covariance matrix, perform eigenvalue decomposition to determine the required dimensionality reduction dimension;

[0145] Based on the dimensionality reduction dimension, determine the dimensionality-reduced feature data.

[0146] In an embodiment of the present application, the secondary radar track prediction model training module is further used for:

[0147] Input the dimension-reduced feature data into the target track prediction module to obtain predicted ADS-B position data;

[0148] Determine the difference based on the predicted ADS-B position data and the corresponding associated ADS-B position data;

[0149] Modify the neuron weights of the target track prediction module based on the difference.

[0150] In an embodiment of the present application, the secondary radar track prediction model training module is further configured to:

[0151] Divide the dimension-reduced feature data and the corresponding associated ADS-B data into a training set and a test set;

[0152] Modify the neuron weights of the target track prediction module based on the effect obtained from the training set, and evaluate the modified secondary radar track prediction model based on the test set.

[0153] In an embodiment of the present application, the secondary radar track prediction model training module is further configured to:

[0154] The model evaluation metrics are root mean square error, correlation coefficient, mean absolute error, and mean bias error. Among them, the calculation formula for root mean square error is:

[0155]

[0156] The calculation formula for correlation coefficient is:

[0157]

[0158] The calculation formula for mean absolute error is:

[0159]

[0160] The calculation formula for mean bias error is:

[0161]

[0162] Among them, y i is the ADS-B azimuth or distance data (true value) of the maneuvering target, and y pred is the target azimuth or distance data (predicted value) predicted by the PCA-LSTM model, is the average of the target true values, m is the total number of samples, and i represents the sample serial number.

[0163] Each module in the above secondary radar track processing device based on the PCA-LSTM algorithm can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0164] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structural diagram can be as Figure 9 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a secondary radar track processing method based on the PCA-LSTM algorithm. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0165] Those skilled in the art can understand that Figure 9 the structure shown in

[0166] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0167] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0167] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0168] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties.

[0170] Those of ordinary skill in the art can understand that all or part of the processes in the methods of 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, database, or other medium used in the embodiments provided in this application 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, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0171] 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.

[0172] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for processing secondary radar tracks based on the PCA-LSTM algorithm, characterized in that The method includes: Obtain historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and ADS-B position data; Train a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, which is used to determine feature data based on the historical secondary radar data and perform dimensionality reduction processing on the feature data using principal component analysis. The target track prediction module is a long short-term memory neural network model based on time series, which is used to predict the future azimuth and distance of the target; Input the secondary radar data of the target to be predicted into the trained secondary radar track prediction model to obtain the target track; The dimensionality reduction module includes: Construct a sample matrix; Perform centering processing on the sample matrix and calculate the covariance matrix of the sample matrix; Perform eigenvalue decomposition based on the covariance matrix to determine the required dimensionality reduction dimension; Determine the dimensionality-reduced feature data based on the dimensionality reduction dimension; The training of the target track prediction module includes: Input the dimensionality-reduced feature data into the target track prediction module to obtain predicted ADS-B position data; Determine the difference based on the predicted ADS-B position data and the corresponding associated ADS-B position data; Modify the neuron weights of the target track prediction module based on the difference.

2. The secondary radar track processing method based on the PCA-LSTM algorithm according to claim 1, wherein The obtaining of the historical secondary radar data and ADS-B position data includes: Based on flight speed and acceleration, obtain the historical secondary radar data and ADS-B position data of low-speed maneuvering targets, medium-speed maneuvering targets, and high-speed maneuvering targets respectively; Based on the distance, obtain the historical secondary radar data and ADS-B position data of nearby targets, medium-nearby targets, medium-far targets, and far targets respectively; Based on the azimuth, obtain the historical secondary radar data and ADS-B position data of targets near the left beam edge, near the beam center, and near the right beam edge respectively.

3. The secondary radar track processing method based on the PCA-LSTM algorithm according to claim 1, characterized in that The associating of the historical secondary radar data and ADS-B position data includes: Associate the historical secondary radar data and ADS-B data of the same target at the same moment according to the target code or S-mode address and time.

4. The secondary radar track processing method based on the PCA-LSTM algorithm according to claim 1, wherein The training of the target track prediction module further includes: Divide the dimensionality-reduced feature data and the corresponding associated ADS-B data into a training set and a test set; Modify the neuron weights of the target track prediction module based on the effect obtained from the training set, and evaluate the modified secondary radar track prediction model based on the test set.

5. The method for processing secondary radar tracks based on the PCA-LSTM algorithm according to claim 4, wherein The evaluating of the modified secondary radar track prediction model based on the test set includes: The model evaluation metrics are root mean square error RMSE, correlation coefficient R2, mean absolute error MAE, and mean bias error MBE. Among them, the calculation formula for the root mean square error is: ， The calculation formula for the correlation coefficient is: ， The calculation formula for the mean absolute error is: ， The calculation formula for the mean bias error is: ， Among them, is the ADS-B azimuth or distance data of the maneuvering target, is the target azimuth or distance data predicted by the PCA-LSTM model, is the average of the target true value, is the total number of samples, represents the sample serial number.

6. A secondary radar track processing device based on the PCA-LSTM algorithm, characterized in that The device includes: A data acquisition module, configured to acquire historical secondary radar data and ADS-B position data, and associate the historical secondary radar data and the ADS-B position data; A secondary radar track prediction model training module, configured to train a secondary radar track prediction model based on the associated historical secondary radar data and ADS-B position data. The secondary radar track prediction model includes a dimensionality reduction module and a target track prediction module. Among them, the dimensionality reduction module is a PCA algorithm model, configured to determine feature data based on the historical secondary radar data, and perform dimensionality reduction processing on the feature data by using principal component analysis. The target track prediction module is a long short-term memory neural network model based on time series, configured to predict the future azimuth and distance of a target; A target track determination module, configured to input the secondary radar data of a target to be predicted into the trained secondary radar track prediction model to obtain a target track; The dimensionality reduction module includes: Construct a sample matrix; Perform centering processing on the sample matrix, and calculate the covariance matrix of the sample matrix; Based on the covariance matrix, perform eigenvalue decomposition to determine the required dimensionality reduction dimension; Based on the dimensionality reduction dimension, determine the dimensionality-reduced feature data; The training of the target track prediction module includes: Input the dimensionality-reduced feature data into the target track prediction module to obtain predicted ADS-B position data; Based on the predicted ADS-B position data and the corresponding associated ADS-B position data, determine a difference; Based on the difference, modify the neuron weights of the target track prediction module.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 5.

Images