Communication positioning scatter point data association method and system based on track prediction

Abstract

The application discloses a communication positioning scattered point data association method and system based on track prediction, wherein the method comprises the following steps: S1, real-time track point data acquisition: reading communication positioning scattered point data HJ from a real-time track point database; S2, track prediction: determining a target category to which a current track point belongs according to a track point target attribute, and selecting a corresponding track prediction model according to the target category, so as to obtain a predicted value of the current track point; and S3, track association: performing track attribute association according to the target attribute of the current track point, and then performing track position association according to a space-time similarity measurement, combining a statistical model and a track prediction result, and obtaining a track association result. According to the application, the track attribute association and the position association are combined with the track prediction value of the track prediction model based on weight combination, so that a more accurate track association result is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of signal fusion, in particular to a communication positioning scatter point data association method and system based on track prediction. BACKGROUND

[0002] In the field of signal fusion, the association and fusion of a large number of real-time generated positioning scatter points are the key and difficulty of situation data fusion, and the fusion results have great guiding significance for subsequent situation behavior prediction and decision implementation. The association of tracks is the key to signal fusion. At present, the track association is mostly based on statistics, gray theory, fuzzy mathematics, and probability multi-hypothesis interconnection methods. However, these methods are not ideal for real-time online data with complex data sources, missing data information, and different characteristics, and are highly dependent on expert experience, or have large computational complexity and slow processing speed. Therefore, how to extract key information from historical tracks to improve the accuracy of track association and processing efficiency is the key to track association research. SUMMARY

[0003] To solve the problem of poor track association caused by complex data sources, missing information, different characteristics, and a large number of association targets, the present application provides a communication positioning scatter point data association method and system based on track prediction, which combines track attribute association, position association, and track prediction values based on a weight combination-based track prediction model to obtain more accurate track association results.

[0004] The technical solution adopted by the present application is as follows:

[0005] A communication positioning scatter point data association method based on track prediction, comprising:

[0006] Step S1. Real-time track point data acquisition: read the communication positioning scatter point data HJ from the real-time track point database;

[0007] Step S2. Track prediction: determine the target category to which the current track point belongs according to the track point target attribute, and select the corresponding track prediction model according to the target category to obtain the prediction value of the current track point;

[0008] Step S3. Track association: perform track attribute association according to the target attribute of the current track point, then perform track position association according to the spatiotemporal similarity measure, combine the statistical model and the track prediction result, and obtain the track association result.

[0009] Further, the process of track prediction comprises:

[0010] Step S21. Determine the target category of the current track point according to the target attribute of the current track point, and the division of the target category is determined by the track prediction model training part;

[0011] Step S22. Select the corresponding track prediction model from the track prediction model library according to the target category;

[0012] Step S23. Obtain the prediction value of the current track point according to the selected track prediction model.

[0013] Further, the flow of the track prediction model training includes:

[0014] Step S221. Divide the target category, and divide the target track in the historical track library according to the target attribute;

[0015] Step S222. Select the binomial interpolation method to interpolate the selected target category data;

[0016] Step S223. Use empirical mode decomposition (EMD) to denoise the interpolated data;

[0017] Step S224. Input the denoised data obtained in step S223 into the weight-based combination prediction model as training data, thereby obtaining a relatively optimal track prediction model.

[0018] Further, the target attribute includes reconnaissance means, target name, country and region, and frequency point.

[0019] Further, the core of the denoising process is to construct an intrinsic mode function (IMF) method and make a judgment based thereon, and the flow of the denoising process includes:

[0020] Step S2231. Input the interpolated data as the original data x(t);

[0021] Step S2232. Initialize the residual data γ0(t) as the original data x(t), and initialize i = 0;

[0022] Step S2233. Let p j-1 (t) = γ i (t), j = 1;

[0023] Step S2234. Calculate the maximum and minimum values of the original data to generate the upper envelope and lower envelope m i (t), n i (t), and calculate the mean values of the upper and lower envelopes:

[0024] avg i (t) = m i (t) + n i (t)

[0025] Step S2235. Calculate the difference between the remaining data and the envelope mean:

[0026] p j (t)=γ i (t)-avg i (t)

[0027] =Step S2236. Calculate the criteria for determining the IMF:

[0028]

[0029] Where T represents the number of sequences;

[0030] Step S2237. Set the threshold parameter ε in advance and determine E. j Does E satisfy? j If the condition is met, proceed to step S2238; otherwise, return to step S2234 to continue processing.

[0031] Step S2238. Let i = i + 1, I i (t)=p j (t), γ i (t)=γ (i-1) (t)-I i (t);

[0032] Step S2239. Determine if the IMF meets the following conditions: (1) For all data, the difference between the number of extreme points and the number of zero points is 0 or 1; (2) For any point, the mean of the envelope determined by the local maximum and local minimum is 0. If the conditions are met, the process ends; otherwise, return to step S2233 to continue processing.

[0033] Furthermore, the processing flow of the weight-based combined prediction model includes:

[0034] Step S2241. Select a recurrent neural network (RNN) as the first single model for trajectory prediction. For a time series dataset containing N data points {x1, x2, ... x...} t …,x N The input source s at the current time t t Including the current input value x t and the inflow information m from the previous moment t-1 The output value includes the predicted value n at time t. t and the information flow d flowing into the next moment t ;

[0035] Step S2242. Select a Long Short-Term Memory Neural Network (LSTM) as the second single model for trajectory prediction;

[0036] Step S2243. Construct a weight-based combined track prediction model, in the form of the following equation:

[0037] y t = ω1 * y 1t + ω2 * y 2t

[0038] where y 1t and y 2t represent the prediction results of the first single model and the second single model at the tth moment, ω1 and ω2 represent the weights, and ω1 + ω2 = 1, 0 ≤ ω1 ≤ 1, and 0 ≤ ω2 ≤ 1.

[0039] Further, the weights of the combined prediction model are determined by using the idea of optimal weighting, and the process includes:

[0040] The error matrix F of the model is calculated:

[0041]

[0042] where m represents the mth prediction moment.

[0043] The error information matrix is:

[0044]

[0045] The error sum of squares of the combined prediction model is:

[0046]

[0047] where W = [ω1, ω2] T , representing the weight value of the combined prediction model, and the optimal weight is the value that minimizes the error sum of squares of the combined prediction model.

[0048] Further, the track association process includes:

[0049] Step S31. Input the current track point HJ t and the jth track in the target track data set L = {l1, l2, …, lh} of the formed track, h represents the number of target tracks in the track data set, and j is initialized to 0. h

[0050] Step S32. Set j = j + 1.

[0051] Step S33. Set the association rule according to the signal parameters and target attributes, and set individualized association rules for data with rich prior information; judge the association target track l j and the current track point HJ t ​If the condition of the attribute association success is met, go to step S34, otherwise, go to step S37;

[0052] Step S34. Calculate the heading angle between the current track point HJ t and the target track l j If the angle is less than the preset threshold r degrees, go to step S35, otherwise, the association fails, go to step S37;

[0053] Step S35. Determine whether the time interval between the current track point HJ t and the target track l j is less than the preset time threshold, if the time interval is less than the time threshold, go to step S36, otherwise, the association fails, go to step S37;

[0054] Step S36. Calculate the distance between the current track point HJ t and the target track l j The distance is calculated by calculating the distance between the track points of the current track point HJ t and the target track l j in the latest time interval, the Euclidean distance is used as the distance metric, and the minimum distance dist min is selected as the distance dist ij between the track point and the target track, the similarity is calculated by using the 3 times error method, the calculation formula is Xsd=exp{(-0.69*dist ij ) / 3*(error1+error2)}, determine whether the similarity is greater than the preset similarity threshold, if the condition is met, go to step S38, otherwise, the association fails, go to step S37;

[0055] Step S37. Determine whether j is less than the number of target tracks h, if the condition is met, return to step S32, otherwise, go to step S38 for track association comprehensive judgment;

[0056] Step S38. Based on the similarity between the preliminary association successful target track and the current track point obtained in step S36, and combined with the predicted value of the target track and the track point at the current time, comprehensively judge which track the current track point should be associated with successfully.

[0057] Further, the flow of the track association comprehensive judgment includes:

[0058] Step S381. If the number of the target tracks associated successfully is d, the similarity between the current track point and the d target tracks constitutes a data set XSD={xsd1,xsd2,…,xsd d};

[0059] Step S382. Calculate the predicted value x of the d target tracks that were initially successfully associated and the track points at the current time. pre,t The Euclidean distance is used to obtain the distance dataset {dist}. 1p ,dist 2p ,…,dist kp ,…,dist dp},dist kp This represents the distance between the k-th target and the predicted value;

[0060] Step S383. Based on the d distances calculated in step S382, set a distance threshold. For distances exceeding the distance threshold, temporarily set their distance to 0. Calculate the mean of the d distances and let it be dist. avg ;

[0061] Step S384. Normalize the d distances respectively. For set elements with non-zero distances, the ε-th point after normalization is: For set elements with a distance of 0, the normalized result is: Thus, the normalized distance dataset ∈={ε1,ε2,…,ε d};

[0062] Step S385. Based on the similarity dataset and normalized distance dataset from steps S381 and S384, subtract the former from the latter at a ratio of σ times to obtain the comprehensive judgment dataset ρ = {ρ1, ρ2, ..., ρ...} d}={xsd1-σ*ε1,xsd2-σ*ε2,…,xsd d -σ*εd, where 0≤σ≤1;

[0063] Step S386. Traverse the dataset ρ, select the largest element, and the target track corresponding to it is the target track that was successfully associated with the current track point.

[0064] A communication positioning scattered data association system based on trajectory prediction includes:

[0065] The real-time waypoint acquisition module is used to read communication positioning scatter data HJ from the real-time waypoint database;

[0066] The trajectory prediction module is used to determine the target category of the current trajectory point based on the target attributes of the trajectory point, select the corresponding trajectory prediction model according to the target category, and thus obtain the predicted value of the current trajectory point;

[0067] The track association module is used to associate track attributes based on the target attributes of the current track point, and then associate track positions based on spatiotemporal similarity measurement, combined with statistical models and track prediction results, to obtain the track association result.

[0068] The application has the beneficial effects that:

[0069] The application obtains more feature information of the track by model training of the historical track, and obtains the prediction value of the real-time track point through the trained prediction model; the track association is performed through the multi-dimensional association analysis of the track attribute such as the signal parameter and the track point prediction value by the space-time similarity measurement, the track feature information is fully utilized, and thus the association accuracy is improved. The online and offline two modules are distributed for processing, the calculation cost is saved, the processing efficiency is improved, and the association accuracy is improved compared with the multi-hypothesis probability interconnection method. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 is a whole model framework diagram.

[0071] Figure 2 is a flowchart of track prediction.

[0072] Figure 3 is a flowchart of track prediction model training.

[0073] Figure 4 is a flowchart of track EMD denoising.

[0074] Figure 5 is a flowchart of the combined prediction model based on weights.

[0075] Figure 6 is a flowchart of track association.

[0076] Figure 7 is a flowchart of track association comprehensive discrimination. DETAILED DESCRIPTION

[0077] In order to have a clearer understanding of the technical features, objectives and effects of the application, the specific embodiments of the application will be described. It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application, that is, the described embodiments are only a part of the embodiments of the application, but not all the embodiments. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0078] Embodiment 1

[0079] Referring to Figure 1 , the embodiment provides a communication positioning scattered point data association method based on track prediction, which comprises the following steps:

[0080] Step S1. Real-time track point data acquisition: reading the communication positioning scattered point data HJ from the real-time track point database;

[0081] Step S2. Track prediction: determining the target category to which the current track point belongs according to the target attribute of the track point, and selecting a corresponding track prediction model according to the target category, so as to obtain the predicted value of the current track point;

[0082] Step S3. Track association: performing track attribute association according to the target attribute of the current track point, and then performing track location association according to the spatiotemporal similarity measure, combining the statistical model and the track prediction result, to obtain the track association result.

[0083] Preferably, referring to Figure 2 , the process of track prediction includes:

[0084] Step S21. Determining the target category to which the track point belongs according to the target attribute of the current track point, and the division of the target category is determined by the track prediction model training part;

[0085] Step S22. Selecting a corresponding track prediction model from the track prediction model library according to the target category;

[0086] Step S23. Obtaining the predicted value of the current track point according to the selected track prediction model.

[0087] Preferably, referring to Figure 3 , the process of track prediction model training includes:

[0088] Step S221. Dividing the target category, and dividing the target track in the historical track library according to the target attribute; preferably, the target attribute includes the reconnaissance means, the target name, the country and region, and the frequency point;

[0089] Step S222. Selecting the binomial interpolation method to perform interpolation processing on the selected target category data;

[0090] Step S223. Using empirical mode decomposition EMD to perform noise reduction processing on the interpolated data;

[0091] Step S224. Inputting the noise-reduced data obtained in step S223 into the weight-based combination prediction model as training data, so as to obtain a relatively optimal track prediction model.

[0092] Preferably, referring to Figure 4 , the core of the noise reduction processing is to construct an intrinsic mode function IMF method and make a judgment according to it, and the process of the noise reduction processing includes:

[0093] Step S2231. Inputting the interpolated data as the original data x(t);

[0094] Step S2232. Initializing the residual data γ0(t) as the original data x(t), and initializing i = 0;

[0095] Step S2233. Let p j-1 (t) = γ i (t), j = 1;

[0096] Step S2234. Calculate the maximum and minimum of the original data, and generate the upper envelope and lower envelope m i (t), n i (t), calculate the mean of the upper and lower envelopes:

[0097] avg i (t) = m i (t) + n i (t)

[0098] Step S2235. Calculate the difference between the remaining data and the envelope mean:

[0099] p j (t) = γ i (t) - avg i (t)

[0100] = Step S2236. Calculate the determination criterion of IMF:

[0101]

[0102] where T represents the number of sequences;

[0103] Step S2237. Set the threshold parameter ε in advance, and judge whether E j satisfies E j < ε. If the condition is met, proceed to step S2238, otherwise, return to step S2234 for further processing;

[0104] Step S2238. Let i = i + 1, I i (t) = p j (t), γ i (t) = γ (i-1) (t) - I i (t);

[0105] Step S2239. Determine whether the IMF satisfies the conditions: (1) for all data, the difference between the number of extreme points and the number of zeros is 0 or 1; (2) for any point, the mean of the envelope determined by the local maximum and the local minimum is 0. If the conditions are met, the process ends, otherwise, return to step S2233 for further processing.

[0106] Preferably, referring to Figure 5 , the processing flow of the combination prediction model based on weights includes:

[0107] Step S2241. Select a recurrent neural network (RNN) as the first single model for trajectory prediction. For a time series dataset containing N data points {x1, x2, ... x...} t …,x N The input source s at the current time t t Including the current input value x t and the inflow information m from the previous moment t-1 The output value includes the predicted value n at time t. t and the information flow d flowing into the next moment t ;

[0108] Step S2242. Select a Long Short-Term Memory Neural Network (LSTM) as the second single model for trajectory prediction;

[0109] Step S2243. Construct a weighted combined trajectory prediction model, expressed as follows:

[0110] y t =ω1*y 1t +ω2*y 2t

[0111] Among them, y 1t y 2t Let ω1 and ω2 represent the prediction results of the first single model and the second single model at time t, respectively. Let ω1 and ω2 represent the weights, and ω1 + ω2 = 1, 0 ≤ ω1 ≤ 1, 0 ≤ ω2 ≤ 1.

[0112] Preferably, the weights of the combined prediction model are determined using the optimal weighting method, and the process includes:

[0113] Find the error matrix F of the model:

[0114]

[0115] Where m represents the m-th prediction time point;

[0116] The error information matrix is ​​as follows:

[0117]

[0118] The sum of squared errors of the combined prediction model is:

[0119]

[0120] Where W = [ω1, ω2] T , represents the weight values ​​of the combined prediction model. The optimal weight is the value that minimizes the sum of squared errors of the combined prediction model.

[0121] Preferably, see Figure 6The flow of track association includes:

[0122] Step S31. Input the current track point HJ t and the jth track in the track data set L of the target track that has been formed, h represents the number of target tracks in the track data set, and j is initialized to 0; h

[0123] Step S32. j = j + 1;

[0124] Step S33. According to the signal parameters and target attributes, set the association rule, and set the personalized association rule for the data with rich prior information; determine whether the attribute association of the target track l j and the current track point HJ t is successful, if the attribute association is successful, go to step S34, otherwise, go to step S37;

[0125] Step S34. Calculate the heading angle of the current track point HJ t and the associated target track l j , if the angle is less than the pre-set threshold r degrees, go to step S35, otherwise, the association fails, and go to step S37;

[0126] Step S35. Determine whether the time interval of the current track point HJ t and the associated target track l j is less than the pre-set time threshold, if the time interval is less than the time threshold, go to step S36, otherwise, the association fails, and go to step S37;

[0127] Step S36. Calculate the distance between the current track point HJ t and the associated target track l j , the calculation method is to calculate the distance between the track points of the current track point HJ t and the target track l j in the nearest period of time, the measurement method is the Euclidean distance, and the minimum distance dist min is selected as the distance dist ij from the track point to the target track, the similarity is calculated by using the 3 times error method, the calculation formula is Xsd = exp{(-0.69*dist ij ) / 3*(error1+error2)}, determine whether the similarity is greater than the pre-set similarity threshold, if the condition is met, go to step S38, otherwise, the association fails, and go to step S37;

[0128] Step S37. Determine whether j is less than the number of target tracks h, if the condition is met, return to step S32, otherwise, go to step S38 for comprehensive judgment of track association; ​

[0129] Step S38. Based on the similarity of the target track and the current track point obtained by the step S36 processing, combined with the predicted value of the target track and the track point at the current time, the current track point should be associated with which track successfully is comprehensively judged.

[0130] Preferably, referring to Figure 7 , the flow of track association comprehensive discrimination includes:

[0131] Step S381. If the number of associated successful target tracks is d, the similarity of the current track point and the d target tracks constitutes the data set XSD={xsd1, xsd2, …, xsd d};

[0132] Step S382. The Euclidean distance of the d target tracks preliminarily associated successfully and the predicted value x pre,t of the track point at the current time is calculated respectively, and the distance data set {dist 1p , dist 2p , …, dist kp , …, dist dp} is obtained, where dist kp represents the distance between the kth target and the predicted value.

[0133] Step S383. Based on the d distances obtained by step S382, a distance threshold is set, and the distance is temporarily set to 0 for the distance exceeding the distance threshold. The average of the d distances is calculated, and is set to dist avg .

[0134] Step S384. The d distances are normalized respectively. For the set elements with distance not equal to 0, the εth normalized point is: For the set elements with distance equal to 0, the normalized point is: Thus, the normalized distance data set ∈={ε1,ε2,…,ε d} is obtained.

[0135] Step S385. Based on the similarity data set and the normalized distance data set of step S381 and step S384, the former and the latter are subtracted in proportion to one-to-one, and the comprehensive judgment data set ρ={ρ1,ρ2,…,ρ d}={xsd1-σ*ε1,xsd2-σ*ε2,…,xsd d -σ*ε d} is obtained, where 0≤σ≤1.

[0136] Step S386. The maximum element is selected by traversing the data set ρ, and the target track corresponding to the maximum element is the last associated successful target track of the current track point.

[0137] Embodiment 2

[0138] This embodiment is based on embodiment 1:

[0139] This embodiment provides a communication positioning scattered point data association system based on track prediction, comprising a real-time track point acquisition module, a track prediction module and a track association module. The real-time track point acquisition module is used to read the communication positioning scattered point data HJ from the real-time track point database. The track prediction module is used to determine the target category to which the current track point belongs according to the track point target attribute, and select the corresponding track prediction model according to the target category, so as to obtain the predicted value of the current track point. The track association module is used to perform track attribute association according to the target attribute of the current track point, and then perform track position association according to the spatio-temporal similarity measure, combined with the statistical model and the track prediction result, to obtain the track association result.

[0140] The processing process of the system includes online processing and offline processing.

[0141] The offline processing is for the track prediction module, which is to process the historical track data. First, a spatio-temporal region is selected, and the data in the region is divided into target categories according to the reconnaissance means, country and region, target attribute, etc. Then, interpolation, noise reduction, training of prediction model, etc. are performed according to the target categories.

[0142] The online processing is for the track association module, which is to process the track points of real-time online communication positioning scattered points. It mainly processes the track association of real-time online communication positioning scattered point data combined with target attribute, spatio-temporal position, expert experience and original track prediction result.

[0143] The track association and the training of the prediction model can be performed simultaneously. The track association is a continuous and uninterrupted processing of real-time data, and the prediction of the prediction model is an intermittent processing according to the interval time and the data volume.

[0144] It should be noted that, for the foregoing method embodiments, in order to simplify the description, they are described as a series of action combinations, but those skilled in the art should know that the present application is not limited to the action order described, because according to the present application, certain steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

Claims

1. A communication positioning scatter point data association method based on track prediction, characterized in that, Comprise: Step S1. Real-time track point data acquisition: read the communication positioning scatter point data from the real-time track point database ; Step S2. Track prediction: determine the target category to which the current track point belongs according to the target attribute of the track point, and select the corresponding track prediction model according to the target category, so as to obtain the predicted value of the current track point; Step S3. Track association: according to the target attribute of the current track point, the track attribute association is carried out, and then according to the spatio-temporal similarity measurement, the track position association is carried out in combination with the statistical model and the track prediction result, so as to obtain the track association result; The flow of the track prediction comprises: Step S21. Determine the target category to which the track point belongs according to the target attribute of the current track point, and the division of the target category is determined by the track prediction model training part; The target attribute comprises a reconnaissance means, a target name, a country and region, and a frequency point; Step S22. Select the corresponding track prediction model from the track prediction model library according to the target category; Step S23. Obtain the predicted value of the current track point according to the selected track prediction model; The flow of the track prediction model training comprises: Step S221. Divide the target category, and divide the target track in the historical track library according to the target attribute; Step S222. Select the binomial interpolation method to interpolate the selected target category data; Step S223. Use empirical mode decomposition (EMD) to denoise the interpolated data; Step S224. Input the denoised data obtained in step S223 into the weight-based combination prediction model as training data, so as to obtain a relatively optimal track prediction model; The flow of the track association comprises: Step S31. Input current track point and the track data set of the target track the jth track in the track data set, denotes the number of target tracks in the track data set, initialize ; Step S32. Let ; Step S33. Set the association rule according to the signal parameter and the target attribute, and set the individualized association rule for the data with rich prior information; judge whether the association target track and the current track point satisfy the condition of attribute association success, if yes, go to step S34, otherwise, go to step S37. and the current track point satisfy the condition of attribute association success, if yes, go to step S34, otherwise, go to step S37; Step S34. Calculate the heading angle between the current track point and the associated target track and the associated target track If the angle is less than a pre-set threshold degree, go to step S35, otherwise, the association fails, go to step S37; Step S35. Determine the current waypoint and associated target tracks If the time interval between the two is less than a preset time threshold, proceed to step S36; otherwise, the association fails and proceed to step S37. Step S36. Calculate the current track point and the distance of the associated target track , which is calculated by calculating the distance between the current track point and the track point of the target track in the latest time period, measured by the Euclidean distance, and selecting the minimum distance as the distance between the track point and the target track , calculating the similarity by the 3 times error method, the formula being , judging whether the similarity is greater than the preset similarity threshold, if the condition is met, turning to step S38, otherwise, the association fails, turning to step S37; Step S37. Determine whether j is less than the target track number h, if the condition is met, return to step S32, otherwise, turn to step S38 for track association comprehensive judgment; Step S38. Based on the similarity of the preliminary association successful target track and the current track point obtained in step S36, in combination with the predicted value of the target track and the track point at the current time, comprehensively judge which track the current track point should be associated with successfully.

2. The track-forecasting-based communication positioning scatter point data association method according to claim 1, characterized in that, The core of the denoising process is to construct an intrinsic mode function (IMF) method and determine it according to the intrinsic mode function (IMF) method. The flow of the denoising process comprises: Step S2231. The interpolated data is input as the original data ; Step S2232. Initialize remaining data for original data , initialize ; Step S2233. Let , ; Step S2234. Calculate the maximum and minimum of the raw data, resulting in an upper envelope and a lower envelope , , Calculate the mean of the upper and lower envelopes: Step S2235. Calculate the difference between the remaining data and the envelope mean value: Step S2236. Calculate the determination criterion of IMF: = wherein T represents the number of sequences; Step S2237. Set threshold parameter in advance Step S2239. Determine whether the IMF meets the conditions: (1) for all data, the difference between the number of extreme points and the number of zero points is 0 or 1; (2) for any point, the mean value of the envelope determined by the local maximum value and the local minimum value is 0; if the condition is met, the process is ended, otherwise, return to step S2233 for continuous processing. , determine whether the condition is met, if the condition is met, proceed to step S2238, otherwise, return to step S2234 for further processing; Step S2238. Let , , t is the time instant; The processing flow of the weight-based combination prediction model comprises:

3. The track-while-predict based communication positioning scatter point data association method according to claim 1, wherein, Step S2242. Select long short-term memory neural network (LSTM) as the second single model for track prediction; Step S2241. Select a recurrent neural network RNN as the first single model for track prediction, for a time series data set containing N data , the input source of the current time includes the current input value and the inflow information of the last time , and the output value includes the predicted value at the time and the information flow flowing into the next time ; Step S2243. Construct a weight-based combination track prediction model, which is expressed as follows: The weight of the combination prediction model is determined by using the optimal weighting method, and the flow comprises: wherein, , denote the prediction results of the first single model and the second single model at the time instant t, respectively, , , denote the weights, and , , .

4. The track-forecasting-based communication positioning scatter point data association method according to claim 3, characterized in that, The error information matrix is: error matrix of the model : wherein, represents the th prediction time point; The error sum of squares of the combination prediction model is: The flow of the track association comprehensive judgment comprises: = wherein, represents the weight value of the combined prediction model, and the optimal weight is the value that minimizes the sum of squared errors of the combined prediction model.

5. The track-forecasting-based communication positioning scatter point data association method according to claim 1, characterized in that, Comprise: Step S381. If the number of successfully associated target tracks is , the similarity of the current track point and target tracks constitutes a data set ; Step S382. Calculate the Euclidean distance between the preliminary successful association target track and the predicted value of the track point at the current time, respectively, to obtain the distance data set The distance between the kth target and the predicted value is represented as dk. ​​​ Step S383. Based on the d distances calculated in step S382, set a distance threshold, temporarily set the distance as 0 for the distance exceeding the distance threshold, calculate the mean value of the d distances, and set it as ; Step S384. Normalize the d distances respectively, and for the set elements with distance not equal to 0, the kth normalized point is: For the set elements with distance equal to 0, the normalized point is: Thus, the normalized distance data set is obtained ; Step S385. Based on the similarity dataset of step S381 and the normalized distance dataset of step S384, the former and the latter are subtracted in proportion to one another - one is twice the other - to obtain a comprehensive decision dataset wherein, ;​ Step S386. Traverse the dataset The largest element is selected, and the target track corresponding to the largest element is the last successful association target track of the current track point.

6. A track-while-predict (TWP) based communication positioning scatter point data association system, applying the communication positioning scatter point data association method as claimed in claim 1, characterized in that, ​ a real-time track point acquisition module, configured to read the communication positioning scattered point data from a real-time track point database ; The track prediction module is configured to determine a target category to which the current track point belongs according to a target attribute of the track point, select a corresponding track prediction model according to the target category, and obtain a predicted value of the current track point. The track association module is configured to perform track attribute association according to the target attribute of the current track point, and then perform track position association according to a time-space similarity measurement, in combination with a statistical model and a track prediction result, to obtain a track association result.

Images