A bearings-only target motion analysis method and system for surface trajectory planning

Abstract

The present invention discloses a pure bearing target motion analysis method and system for surface track planning; wherein the analysis method comprises: a preprocessing step, a target predicted position acquisition step, and a target predicted position correction value acquisition step; the present invention uses a target position prediction model based on a time convolutional network to obtain a target predicted position, and compared with a traditional positioning method, the target position prediction model has a higher prediction accuracy; the present invention uses a set empirical mode decomposition method to decompose an initial signal, and uses a target position prediction model to obtain a target predicted position, thereby not prone to problems of reduced solution accuracy or even divergence; the present invention constructs a motion state equation of a target, and on this basis uses a recursive least squares method to correct the target predicted position output by the target position prediction model, so that all observation information in the observation process is utilized, and the accuracy and stability of the target motion state prediction are further improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of target motion analysis for track planning, and more specifically, relates to a bearing-only target motion analysis method and system for surface track planning. Background Art

[0002] The track planning technology of underwater unmanned submersibles in complex sea conditions is an important research direction in marine science, and plays a pivotal role in both military and civilian fields. Target positioning is one of the key technologies for underwater unmanned submersible track planning. Accurate positioning of obstacles in the navigation path is the basis for making intelligent agent optimization decisions and planning track tasks. Pure bearing target motion analysis obtains the motion state of obstacles such as position and speed by observing and analyzing the obstacle bearing information. It has the advantages of long range, high flexibility, and strong concealment, and is currently a widely used positioning method.

[0003] In the process of pure bearing target motion analysis, the measurement party needs to continuously obtain the azimuth of the target through the observation equation. However, traditional pure target motion analysis methods such as the extended Kalman filter algorithm and the unscented Kalman filter algorithm are affected by the nonlinear characteristics of the observation equation, and are prone to problems of reduced solution accuracy or even divergence. Using deep learning methods for pure target motion analysis can automatically learn representative features from data and establish a nonlinear mapping relationship between input features and output predictions. It has been successfully applied in the field of time series prediction, but there are still the following shortcomings: 1) The input dimension of the deep learning network is limited, and it can only process part of the observation information, and the information utilization rate is insufficient; 2) In the problem of pure bearing target motion analysis, it is assumed that the target maintains uniform linear motion. However, traditional deep learning methods have not yet considered this dynamic constraint, resulting in large fluctuations in the prediction results.

[0004] Therefore, the decline in solution accuracy, insufficient utilization of observation information, and large fluctuations in prediction results in the existing technology are problems that need to be solved urgently.

[0005] In order to facilitate the understanding of the present invention, the following explains relevant terms and related concepts:

[0006] Temporal Convolutional Network: A convolutional neural network structure used to process time series data; its principle is to introduce causal convolution, dilated convolution, residual module and other mechanisms on the basis of convolutional neural network to expand the receptive field, reduce computational complexity and improve network performance.

[0007] Gradient descent method: an optimization algorithm; often used in deep learning to recursively approximate the minimum deviation model; its principle is to repeatedly calculate the gradient so that the loss function continues to decrease in the direction of the local minimum.

[0008] Ensemble empirical mode decomposition method: a noise-assisted signal decomposition method; its principle is to use the statistical characteristics of Gaussian white noise with uniform frequency distribution, and change the extreme point characteristics of the signal by adding different white noises of the same amplitude each time, and then perform overall averaging on the corresponding sub-signals obtained by multiple empirical mode decompositions to offset the added white noise, thereby effectively suppressing the generation of modal aliasing;

[0009] Recursive least squares method: a method for estimating the current real output based on the data generated by the system's past operation; its principle is to use the least squares method to estimate the current real output of the system based on the latest measurement results in a recursive manner;

[0010] Minimum and maximum normalization method: a data preprocessing method commonly used in deep learning. By scaling the attribute data so that it falls into a small specific range, the attributes of the data can be further analyzed.

[0011] Sampling point: The observer adopts a "Z"-shaped observation path, and each target or sample target maintains uniform linear motion; a two-dimensional plane rectangular coordinate system is established with due north as the positive direction of the y-axis and due east as the positive direction of the x-axis, taking any point within 500 meters of the observer's initial position as the origin; the coordinates of the observer's initial position are (0, 100), and the speed in the x-axis direction and the speed in the y-axis direction are 0m / s and 20m / s respectively; the coordinates of the initial position of a single target or sample target (x 0 ,y 0 ) is in the range of x 0 ,y 0 ∈[1500m,2500m], speed v in the x-axis direction x and the velocity v in the y-axis direction y The range is v x ,v y ∈[8m / s,12m / s]; within the preset measurement time, the observer uses the passive detection radar to measure the azimuth between each target or sample target and the observer multiple times; wherein the preset measurement time is greater than 500 seconds, and the number of measurements is greater than 50 times; each measurement of each target or sample target obtains a sampling point, which is composed of the observer coordinates at the time of measurement and the azimuth of the target or sample target;

[0012] Target initial motion state: When a single target is at the initial position, the target's coordinates, the target's speed in the x-axis direction, and the target's speed in the y-axis direction constitute the target's initial motion state;

[0013] Target actual motion state: In a single measurement, the coordinates of a single target or sample target, the speed of the target or sample target in the x-axis direction and the speed in the y-axis direction constitute the target actual motion state;

[0014] Target predicted position: the predicted value of the coordinates of a single target or sample target;

[0015] Target location: the actual value of the coordinates of a single target or sample targets; Summary of the invention

[0016] In view of the defects of the prior art, the purpose of the present invention is to provide a pure bearing target motion analysis method and system for surface trajectory planning, aiming to solve the problems of reduced solution accuracy, insufficient utilization of observation information and large fluctuations in prediction results in the prior art.

[0017] To achieve the above object, the present invention provides a method for obtaining a target position prediction model, comprising the following steps:

[0018] (1) Sub-signal acquisition steps:

[0019] The original signal of each sample target is decomposed by using the ensemble empirical mode decomposition method to obtain multiple original sub-signals;

[0020] The single original signal is a discrete curve composed of various sampling points obtained by performing multiple measurements on a single sample target;

[0021] (2) Dataset division steps:

[0022] The original sub-signal is standardized by using the minimum-maximum standardization method to obtain a standardized signal; the standardized signal of each sample target and the actual motion state of the sample target constitute a sample; all samples constitute a data set, and the data set is divided into a training set and a test set;

[0023] (3) Model acquisition steps:

[0024] The temporal convolutional network is trained multiple times: each training takes the training set as input, the target predicted position of each sample target as output, and the gradient descent method is used to reduce the difference between the target position of each sample target and the target predicted position; each training obtains a model;

[0025] The performance of each model is evaluated by the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

[0026] Furthermore, in the model acquisition step, the temporal convolutional network includes three convolutional layers and one fully connected layer, and the input dimension of the temporal convolutional network is 10;

[0027] The present invention provides a bearing-only target motion analysis method based on the target position prediction model as described above, comprising the following steps:

[0028] (1) Preprocessing steps:

[0029] The initial signal of each target is decomposed by using the ensemble empirical mode decomposition method to obtain multiple sub-signals; the single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target;

[0030] Performing a standardization operation on the sub-signal using a minimum-maximum standardization method;

[0031] (2) Steps for obtaining the target predicted position:

[0032] Inputting the sub-signals after the normalization operation into the target position prediction model to obtain the target prediction position of each target;

[0033] (3) Steps for obtaining target predicted position correction value:

[0034] The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement;

[0035] The motion state equation is:

[0036] Y(k)=Φ(k)θ+e(k)

[0037] Wherein, k is the kth measurement of each target, k>0; Y(k) is the predicted position of the target, and Corresponding to the predicted values ​​of the target's coordinates in the x-axis direction and the y-axis direction; Φ(k) is the parameter matrix, and Φ(k) = [k 1]; e(k) is the error vector; θ is the initial motion state of the target;

[0038] The target predicted position correction value is:

[0039]

[0040] Where Z(k) is the target predicted position correction value of the kth measurement; is the estimated value of the initial motion state of the target measured for the kth time.

[0041] Furthermore, in the target predicted position correction value acquisition step, the process of using the recursive least squares method to obtain the estimated value of the target initial motion state is:

[0042]

[0043] K(k)=P(k)Φ(k),

[0044]

[0045] Where, P(k) is the covariance matrix; K(k) is the corrected gain matrix; is the estimated value of the initial motion state of the target measured for the kth time.

[0046] The present invention provides a system for acquiring a target position prediction model, comprising the following modules:

[0047] Sub-signal acquisition module:

[0048] The original signal of each sample target is decomposed by using an ensemble empirical mode decomposition system to obtain multiple original sub-signals;

[0049] The single original signal is a discrete curve composed of various sampling points obtained by performing multiple measurements on a single sample target;

[0050] Dataset partitioning module:

[0051] The original sub-signal is standardized by using the minimum-maximum standardization method to obtain a standardized signal; the standardized signal of each sample target and the actual motion state of the sample target constitute a sample; all samples constitute a data set, and the data set is divided into a training set and a test set;

[0052] Model acquisition module:

[0053] The temporal convolutional network is trained multiple times: each training takes the training set as input, the target predicted position of each sample target as output, and the gradient descent method is used to reduce the difference between the target position of each sample target and the target predicted position; each training obtains a model;

[0054] The performance of each model is evaluated by the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

[0055] Furthermore, in the model acquisition module, the temporal convolutional network includes three convolutional layers and one fully connected layer, and the input dimension of the temporal convolutional network is 10;

[0056] The present invention provides a bearing-only target motion analysis system based on the target position prediction model described above, comprising the following modules:

[0057] Preprocessing module:

[0058] An ensemble empirical mode decomposition system is used to decompose the initial signal of each target to obtain multiple sub-signals; the single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target;

[0059] Performing a standardization operation on the sub-signal using a minimum-maximum standardization method;

[0060] Target prediction position acquisition module:

[0061] Inputting the sub-signals after the normalization operation into the target position prediction model to obtain the target prediction position of each target;

[0062] Target predicted position correction value acquisition module:

[0063] The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement;

[0064] The motion state equation is:

[0065] Y(k)=Φ(k)θ+e(k)

[0066] Wherein, k is the kth measurement of each target, k>0; Y(k) is the predicted position of the target, and Corresponding to the predicted values ​​of the target's coordinates in the x-axis direction and the y-axis direction; Φ(k) is the parameter matrix, and Φ(k) = [k 1]; e(k) is the error vector; θ is the initial motion state of the target;

[0067] The target predicted position correction value is:

[0068]

[0069] Where Z(k) is the target predicted position correction value of the kth measurement; is the estimated value of the initial motion state of the target measured for the kth time.

[0070] Furthermore, in the target predicted position correction value acquisition module, the process of using the recursive least squares method to obtain the estimated value of the target initial motion state is:

[0071]

[0072] K(k)=P(k)Φ(k),

[0073]

[0074] Where, P(k) is the covariance matrix; K(k) is the corrected gain matrix; is the estimated value of the initial motion state of the target measured for the kth time.

[0075] The present invention provides a pure bearing target motion analysis device, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to implement the pure bearing target motion analysis method as described above when executing the computer program.

[0076] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned pure bearing target motion analysis method is implemented.

[0077] Compared with the prior art, the above technical scheme conceived by the present invention uses the ensemble empirical mode decomposition method to decompose the initial signal, and uses the target position prediction model based on the time convolution network to obtain the target predicted position, so that the nonlinear mapping established by the observation equation is more reliable and is less likely to have the problem of reduced solution accuracy or even divergence.

[0078] The present invention constructs the motion state equation of the target, and on this basis uses the recursive least squares method to correct the target predicted position output by the target position prediction model, so that all observation information in the observation process is utilized, and the final target position is more in line with the characteristics of the target uniform linear motion than before correction, further improving the accuracy and stability of the target motion state prediction.

[0079] The present invention uses a collective empirical mode decomposition method to decompose the initial signal, so that the decomposed sub-signals can effectively retain the initial signal information while improving the feature dimension.

[0080] The present invention uses a target position prediction model based on a temporal convolutional network to obtain a target predicted position. Compared with traditional positioning methods, the target position prediction model has a higher prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 is a flow chart of a method for acquiring a target position prediction model provided by an embodiment of the present invention;

[0082] Figure 2 A structural diagram of a temporal convolutional network provided by an embodiment of the present invention;

[0083] Figure 3 is a flow chart of a bearing-only target motion analysis method provided by an embodiment of the present invention;

[0084] Figure 4 A schematic diagram of multiple sub-signals obtained by decomposing an initial signal using an ensemble empirical mode decomposition method provided in an embodiment of the present invention;

[0085] Figure 5A schematic diagram showing a comparison between the average values ​​of the position root mean square error of the bearing-only target motion analysis method provided by an embodiment of the present invention and that of the prior art;

[0086] Figure 6 A schematic diagram showing a comparison between the average values ​​of the predicted velocity root mean square error of the bearing-only target motion analysis method provided in an embodiment of the present invention and that of the prior art. DETAILED DESCRIPTION

[0087] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0088] like Figure 1 As shown, an embodiment of the present invention provides a method for obtaining a target position prediction model, comprising the following steps:

[0089] (1) Sub-signal acquisition steps:

[0090] The original signal of each sample target is decomposed by using the ensemble empirical mode decomposition method to obtain multiple original sub-signals;

[0091] A single original signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single sample target; the number of measurements is 70 times, and the sampling interval between each sampling point is 10 seconds.

[0092] (2) Dataset division steps:

[0093] The minimum and maximum value normalization method is used to normalize the original sub-signals to obtain standardized signals; the standardized signal of each sample target and the actual motion state of the sample target constitute a sample; all samples constitute a data set, which is divided into a training set and a test set; the data set contains 10,000 samples, and the ratio of the training set to the test set is 9:1.

[0094] (3) Model acquisition steps:

[0095] The temporal convolutional network was trained five times, with the training set as input for each training: the standardized signal in each sample in the training set was evenly divided into 7 subsequences, and these 7 subsequences were input into the temporal convolutional network in sequence; the target predicted position of each sample target was used as output; the gradient descent method was used to reduce the difference between the target position and the target predicted position of each sample target; a model was obtained after each training;

[0096] like Figure 2As shown in the figure, the temporal convolutional network contains three convolutional layers and one fully connected layer. The three convolutional layers are convolutional layer 1, convolutional layer 2, and convolutional layer 3. The input dimension of the temporal convolutional network is 10. The main parameters of the temporal convolutional network are shown in the following table:

[0097]

[0098] The performance of each model is evaluated through the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

[0099] like Figure 3 As shown, an embodiment of the present invention provides a bearing-only target motion analysis method based on the target position prediction model as described above, comprising the following steps:

[0100] (1) Preprocessing steps:

[0101] like Figure 4 As shown, the ensemble empirical mode decomposition method is used to decompose the initial signal of each target to obtain three sub-signals, namely sub-signal 1, sub-signal 2, and sub-signal 3; a single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target;

[0102] The sub-signals are standardized using the minimum and maximum value normalization method;

[0103] (2) Steps for obtaining the target predicted position:

[0104] The sub-signals after the normalization operation are input into the target position prediction model to obtain the target prediction position of each target;

[0105] (3) Steps for obtaining target predicted position correction value:

[0106] The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement;

[0107] The state equation of motion is:

[0108] Y(k)=Φ(k)θ+e(k)

[0109] Where k is the kth measurement of each target, k>0; Y(k) is the predicted position of the target, and Corresponding to the predicted values ​​of the target's coordinates in the x-axis direction and the y-axis direction; Φ(k) is the parameter matrix, and Φ(k) = [k 1]; e(k) is the error vector; θ is the initial motion state of the target;

[0110] The process of using the recursive least squares method to obtain the estimated value of the initial motion state of the target is:

[0111]

[0112] K(k)=P(k)Φ(k),

[0113]

[0114] Where, P(k) is the covariance matrix; K(k) is the corrected gain matrix; is the estimated value of the initial motion state of the target measured at the kth time;

[0115] The target predicted position correction value is:

[0116]

[0117] Where Z(k) is the target predicted position correction value of the kth measurement; is the estimated value of the initial motion state of the target measured for the kth time.

[0118] The root mean square error RMSE position RMSE of the predicted speed velocity Used to evaluate the difference between the actual motion state of the target and the predicted value; RMSE position and RMSE velocity They are:

[0119]

[0120]

[0121] Among them, (x, y) is the target position, v is the target predicted position correction value; x and v y are the actual values ​​of the target's speed in the x-axis direction and the speed in the y-axis direction, and Corresponding to v x and v y The predicted value of .

[0122] As attached Figure 5 and Figure 6 As shown, after 50 Monte Carlo experiments, the average values ​​of the root mean square errors of the positions obtained by the extended Kalman filter (EKF), the unscented Kalman filter (UKF) and the present invention are 2273.5, 176.2 and 120.8 respectively; the average values ​​of the root mean square errors of the predicted speeds are 6.34, 0.32 and 0.26 respectively; the results show that compared with the prior art, the present invention has better prediction performance.

[0123] The embodiment of the present invention provides a system for acquiring a target position prediction model, comprising the following modules:

[0124] Sub-signal acquisition module:

[0125] The original signal of each sample target is decomposed by using an ensemble empirical mode decomposition system to obtain multiple original sub-signals;

[0126] A single raw signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single sample target;

[0127] Dataset partitioning module:

[0128] The original sub-signal is standardized by the minimum-maximum standardization method to obtain a standardized signal; the standardized signal of each sample target and the actual motion state of the sample target form a sample; all samples form a data set, which is divided into a training set and a test set;

[0129] Model acquisition module:

[0130] The temporal convolutional network is trained multiple times: each training takes the training set as input and the target predicted position of each sample target as output, and the gradient descent method is used to reduce the difference between the target position of each sample target and the target predicted position; each training obtains a model;

[0131] The temporal convolutional network contains three convolutional layers and one fully connected layer, and the input dimension of the temporal convolutional network is 10;

[0132] The performance of each model is evaluated through the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

[0133] The embodiment of the present invention provides a bearing-only target motion analysis system based on the target position prediction model as described above, comprising the following modules:

[0134] Preprocessing module:

[0135] The initial signal of each target is decomposed by using the ensemble empirical mode decomposition system to obtain multiple sub-signals; a single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target;

[0136] The sub-signals are standardized using the minimum and maximum value normalization method;

[0137] Target prediction position acquisition module:

[0138] The sub-signals after the normalization operation are input into the target position prediction model to obtain the target prediction position of each target;

[0139] Target predicted position correction value acquisition module:

[0140] The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement;

[0141] The state equation of motion is:

[0142] Y(k)=Φ(k)θ+e(k)

[0143] Where k is the kth measurement of each target, k>0; Y(k) is the predicted position of the target, and Corresponding to the predicted values ​​of the target's coordinates in the x-axis direction and the y-axis direction; Φ(k) is the parameter matrix, and Φ(k) = [k 1]; e(k) is the error vector; θ is the initial motion state of the target;

[0144] The process of using the recursive least squares method to obtain the estimated value of the initial motion state of the target is:

[0145]

[0146] K(k)=P(k)Φ(k),

[0147]

[0148] Where, P(k) is the covariance matrix; K(k) is the corrected gain matrix; is the estimated value of the initial motion state of the target measured for the kth time.

[0149] The target predicted position correction value is:

[0150]

[0151] Where Z(k) is the target predicted position correction value of the kth measurement; is the estimated value of the initial motion state of the target measured for the kth time.

[0152] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A bearings-only target motion analysis method for surface track planning, characterized in that: The following steps are involved: (1) Preprocessing steps: The ensemble empirical mode decomposition method is used to decompose the initial signal of each target to obtain multiple sub-signals; a single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target; Performing a standardization operation on the sub-signal using a minimum-maximum standardization method; (2) Steps for obtaining the predicted target position: The sub-signals after the normalization operation are input into the target position prediction model to obtain the target prediction position of each target; (3) Steps for obtaining the target predicted position correction value: The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement; The motion state equation is: in, k For each target k Measurements, k > 0; Y(k) predicting a position for the target, , and Corresponding to the predicted values ​​of the target's coordinates in the x-axis and y-axis directions; is a parameter matrix, and ; e ( k ) is the error vector; is the initial motion state of the target; The target predicted position correction value is: in, For the k The target predicted position correction value measured this time; For the k The estimated value of the initial motion state of the target measured by this method; The method for acquiring the target position prediction model includes: The original signal of each sample target is decomposed by using the ensemble empirical mode decomposition method to obtain multiple original sub-signals; The single original signal is a discrete curve composed of various sampling points obtained by performing multiple measurements on a single sample target; The original sub-signal is standardized by using the minimum-maximum standardization method to obtain a standardized signal; the standardized signal of each sample target and the actual motion state of the sample target constitute a sample; all samples constitute a data set, and the data set is divided into a training set and a test set; The temporal convolutional network is trained multiple times: each training takes the training set as input, the target predicted position of each sample target as output, and the gradient descent method is used to reduce the difference between the target position of each sample target and the target predicted position; each training obtains a model; The performance of each model is evaluated by the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

2. The method for bearings-only target motion analysis according to claim 1, characterized in that: In the model acquisition step, the temporal convolutional network includes three convolutional layers and one fully connected layer, and the input dimension of the temporal convolutional network is 10.

3. The method for bearings-only target motion analysis according to claim 1, characterized in that: In the target predicted position correction value acquisition step, the process of using the recursive least squares method to obtain the estimated value of the target initial motion state is as follows: , , , in, P ( k ) covariance matrix; K ( k ) is the modified gain matrix; For the k The estimated value of the initial motion state of the target measured.

4. A bearings-only target motion analysis system for surface track planning, characterized in that: Includes the following modules: Preprocessing module: The initial signal of each target is decomposed by using the ensemble empirical mode decomposition system to obtain multiple sub-signals; a single initial signal is a discrete curve composed of various sampling points obtained by multiple measurements of a single target; Performing a standardization operation on the sub-signal using a minimum-maximum standardization method; Target prediction position acquisition module: The sub-signals after the normalization operation are input into the target position prediction model to obtain the target prediction position of each target; Target predicted position correction value acquisition module: The recursive least square method is used to estimate the initial motion state of the target in the motion state equation, and the estimated value of the initial motion state of the target in a single measurement is obtained, thereby obtaining the correction value of the target predicted position in a single measurement; The motion state equation is: in, k For each target k Measurements, k > 0; Y(k) predicting a position for the target, , and Corresponding to the predicted values ​​of the target's coordinates in the x-axis and y-axis directions; is a parameter matrix, and ; e ( k ) is the error vector; is the initial motion state of the target; The target predicted position correction value is: in, For the k The target predicted position correction value measured this time; For the k The estimated value of the initial motion state of the target measured by this method; The method for acquiring the target position prediction model includes: The original signal of each sample target is decomposed by using the ensemble empirical mode decomposition method to obtain multiple original sub-signals; The single original signal is a discrete curve composed of various sampling points obtained by performing multiple measurements on a single sample target; The original sub-signal is standardized by using the minimum-maximum standardization method to obtain a standardized signal; the standardized signal of each sample target and the actual motion state of the sample target constitute a sample; all samples constitute a data set, and the data set is divided into a training set and a test set; The temporal convolutional network is trained multiple times: each training takes the training set as input, the target predicted position of each sample target as output, and the gradient descent method is used to reduce the difference between the target position of each sample target and the target predicted position; each training obtains a model; The performance of each model is evaluated by the test set: the model with the smallest difference between the target position and the target predicted position is the target position prediction model.

5. The bearing-only target motion analysis system according to claim 4, characterized in that: In the model acquisition module, the temporal convolutional network includes three convolutional layers and one fully connected layer, and the input dimension of the temporal convolutional network is 10.

6. The bearing-only target motion analysis system according to claim 4, characterized in that: In the target predicted position correction value acquisition module, the process of using the recursive least squares method to obtain the estimated value of the target initial motion state is as follows: , , , in, P ( k ) covariance matrix; K ( k ) is the modified gain matrix; For the k The estimated value of the initial motion state of the target measured.

7. A bearing-only target motion analysis device, characterized in that: It comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement the pure bearing target motion analysis method as described in any one of claims 1 to 3 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the bearing-only target motion analysis method as described in any one of claims 1 to 3 is implemented.

Images