CN113341379A

CN113341379A - Radar signal sorting method based on adaptive threshold and iterative control

Info

Publication number: CN113341379A
Application number: CN202110197821.9A
Authority: CN
Inventors: 傅雄军; 王晓妍; 董健; 谢民; 卢继华; 杜慧茜
Original assignee: Beijing Institute of Technology BIT; Army Engineering University of PLA
Current assignee: Beijing Institute of Technology BIT; Army Engineering University of PLA
Priority date: 2021-02-22
Filing date: 2021-02-22
Publication date: 2021-09-03

Abstract

The invention relates to a radar signal sorting method based on adaptive threshold and iterative control, in particular to a radar signal sorting method based on adaptive threshold and iterative control PSO and k-means, and belongs to the technical field of population evolution and signal classification. The method comprises the following steps: preprocessing radar pulse PDW stream data and generating effective cluster number; primarily sorting radar signals by using a Particle Swarm Optimization (PSO) method with adaptive parameter change and iterative control; and finally, performing data clustering enhancement on the radar signal by using k-means. By overcoming the defects of the k-means and PSO methods, the method can automatically determine the clustering number, find a globally optimal clustering center, and has the advantages of high convergence rate, less iteration times, high sorting recognition rate and capability of improving various clustering evaluation indexes.

Description

Radar signal sorting method based on adaptive threshold and iterative control

Technical Field

The application relates to a radar signal sorting method based on adaptive threshold and iterative control, in particular to a radar signal sorting method based on adaptive threshold and iterative control PSO and k-means, and belongs to the technical field of population evolution and signal classification.

Background

With the progress and development of radar technology, the number of radiation sources in an electromagnetic space is large, the density is high, the signal modulation is complex, the distribution is wide, and meanwhile, the signal overlapping is serious, so that the sorting work of the radar radiation sources is difficult. The traditional sorting method is stranded in the face of increasingly complex electromagnetic environments, so that the real-time and effective sorting of high-density complex radar signals is a key problem to be solved by the current radar detection system.

The k-means method is a traditional clustering method based on partitioning. The method has the advantages of simplicity and high convergence rate, so that the method is more generally applied. It has several drawbacks: (1) the number of clusters has to be predetermined. (2) The random k center points affect the clustering effect. (3) The k-means clustering method is easy to fall into local optimization. Many improved methods based on k-means have emerged. Some scholars propose a k-means clustering radar signal sorting method based on a data field, and the method is used for removing noise points and determining an initial clustering center according to a data field theory; the scholars also define the local density by adopting the k-means idea and design a new cluster center selection by utilizing the distance normalization principle; and the learner also selects k initial clustering centers for improvement by adopting a relationship matrix and dimensionality centrality, so that a certain effect is achieved.

In recent years, with the development of optimization methods, many scholars have studied the application of some heuristic optimization methods, such as genetic methods, particle swarm optimization methods, ant colony optimization methods, simulated annealing methods, and the like. They take advantage of these optimization methods to ameliorate some of the disadvantages of conventional clustering methods. The PSO method is an intelligent optimization method which is widely applied at present and is used for searching global optimum based on population, has high dispersity, particle-to-particle collaboration and simple implementation, and is suitable for a plurality of optimization problems. One of the main application areas of the method is cluster analysis. Many studies have adopted the PSO method to accomplish the task of clustering data, like the FCM method based on PSO optimization. But the PSO requires more iterations than k-means in order to converge to a globally optimal solution. And the parameters involved in the method have a great influence on the sorting result. Therefore, in order to obtain an effective clustering result, it is necessary to adopt the two methods to be correspondingly improved and effectively combined.

The present application addresses the shortcomings of the k-means and PSO methods described above, proposing a k-means method based on contour coefficients and an improved method based on a combination. Experiments prove that the method has advantages in the aspects of recognition rate, various clustering evaluation indexes, iteration times and the like compared with other optimization calculations.

Disclosure of Invention

The method aims to solve the technical problems that the existing radar sorting method based on k-means and particle swarm optimization is indefinite in clustering number, randomized in initial clustering center, easy to fall into local optimization, relatively slow in PSO method convergence time and greatly influenced by preset parameters in sorting results, and provides a radar signal sorting method based on adaptive threshold and iterative control.

In order to achieve the purpose, the following technical scheme is adopted in the application.

The radar signal sorting method comprises the following steps: the method comprises the following steps of radar pulse PDW stream data preprocessing, radar signal primary sorting and radar signal enhanced data clustering:

step 1, preprocessing radar pulse PDW stream data to generate an effective clustering number;

preprocessing radar pulse PDW stream data, wherein the preprocessing comprises constructing a radar data set, collecting and revising data to be sorted and evaluating the effective cluster number of clustering based on the contour coefficient;

step 1, specifically comprising the following substeps:

step 1.1, constructing a radar data set;

the dimension of the constructed radar data set is D, the radar data set comprises data to be sorted formed by generating different types of radar signals, and each type of radar signal is a group of PDW flow data;

step 1.2, collecting data to be sorted from the radar data set constructed in step 1.1;

step 1.3, revising the data to be sorted, namely changing different pulse signal data in the data to be sorted into the same scale range, and generating revised data to be sorted;

the maximum value of the data to be sorted is recorded as Mmax, and the minimum value of the data to be sorted is recorded as Mmin;

step 1.4, calculating a contour coefficient, and evaluating the effective cluster number of the cluster based on the contour coefficient, wherein the method specifically comprises the following steps:

step 1.4.1, calculating a contour coefficient;

step 1.4.2, evaluating the effective cluster number of the clusters based on the contour coefficient, namely selecting the cluster number with the maximum contour coefficient as the optimal cluster number, and specifically: giving a range of the number of clusters by adopting an exhaustion method, calculating and comparing profile coefficients in the range, and selecting the number of clusters corresponding to the maximum profile coefficient as the optimal cluster number;

wherein, the optimal clustering number is marked as M;

step 2, primarily sorting the radar signals to generate a clustering center after primary sorting, which specifically comprises the following steps: searching a revised global optimal solution of the data to be sorted by adopting an improved particle swarm optimization method based on adaptive parameter change and iterative control, wherein the global optimal solution is a data clustering center after primary sorting;

step 2, comprising the following substeps:

step 2.1, initializing each parameter of the particle swarm optimization method;

wherein, each parameter comprises the population number, the maximum iteration number, the particle position, the particle speed, the maximum movement speed of the particle and the historical highest fitness function value of the particle;

the population number is recorded as N, and the value range of N is 10 to 50; the value range of the maximum iteration times is between 50 and 200 and is marked as Tmax; the positions of the particles are position matrixes of M multiplied by D multiplied by N dimensions, and elements in the position matrixes are random numbers between Mmin and Mmax; the particle velocity is a velocity matrix with dimension of M multiplied by D multiplied by N, and each element in the velocity matrix is a random number with the value in the range of 0-1; and the maximum value in the velocity matrix is between 0.5-1.0; the historical highest fitness function value of the particle is initialized to infinity;

step 2.2, determining the range of the inertia weight and the acceleration coefficient;

wherein, the inertia weight is marked as w, and the value range is between 0.5 and 1.2; the acceleration coefficient is marked as c, and the value range of the acceleration coefficient is between 0 and 4;

step 2.3, setting the time t to be 1, and taking the initial values of the position and the speed of the particle swarm as the position and the speed of the particle swarm at the time 0;

step 2.4, calculating a fitness function value of each particle at the time t, specifically: calculating the position of the particle obtained by the particle swarm method at the time t-1, namely the Euclidean distance between the clustering center and the corresponding reference clustering center, thereby being used as the mean square error value between the clustering center and the reference clustering center, namely the fitness function value;

when t is 1, the position of the particle at the time t-1, namely at the time 0, is the initialized particle position;

step 2.5, calculating the optimal position pbest in the particle at the current time t and the global optimal value gbest of the particle swarm, specifically:

step 2.5.1, calculating the optimal position pbest of each particle, namely selecting the position which enables the fitness function value of the particle to be minimum as the optimal position; for each particle, obtaining a fitness function value at the time t according to the step 2.4, comparing the fitness function value with the historical highest fitness function value at the time t-1 which the particle has been subjected to, if the fitness function value is better, namely the fitness function value is smaller, updating the optimal position of the particle, if the fitness function value is not better than the previous fitness function value, not updating the optimal position of the particle, and continuously keeping the optimal position which the last time, namely the time t-1 has been subjected to;

when t is 1, the historical highest fitness function value of the particle at the time t-1, namely the time 0, is the initialized historical highest fitness function value;

step 2.5.2, calculating a global optimal position gbest in the particle swarm, namely comparing the fitness function value corresponding to the optimal position of each particle obtained in the step 2.5.1, selecting the position with the minimum corresponding fitness function value as the optimal position in the whole particle swarm, and recording the found minimum fitness function value as the global optimal fitness function value of the particle swarm;

step 2.6, updating the inertia weight w and the acceleration coefficient c of the particle swarm by adopting a linear increasing mode from the minimum value to the maximum value;

step 2.7, updating the speed and position of the particle swarm to generate a new particle swarm, which specifically comprises the following steps:

2.7A speed of updating the particle swarm is specifically as follows: adding the inertial memory term, the self-cognition term and the group cognition term of the particle;

wherein, the inertia memory term is obtained by multiplying the inertia weight coefficient w by the speed of the particles at the time of t-1; the self-identification item is obtained by multiplying a vector which is obtained by calculating the best position point pbest of the particle pointed by the particle position point at the moment t-1 by an acceleration coefficient and a random number in the range of 0 to 1; the group recognition item is obtained by multiplying a vector obtained by calculating the best position point gbest of the particle position point pointing to the group at the time t-1 by an acceleration coefficient and a random number in a range from 0 to 1;

if the updating speed of the particles is greater than the maximum speed of the particles obtained in the step 2.1, the updating speed at the current time t is taken as a maximum speed value; if the updating speed of the particles is less than the negative value of the maximum speed of the particles set in the step 2.1, taking the updating speed at the current time t as the negative value of the maximum speed value;

2.7B updating the position of the particle swarm, specifically: adding the position of the particle at the t-1 moment and the speed of the particle at the current t moment;

if the position of the particle is larger than the maximum value Mmax of the position, the maximum value is taken as the position of the particle at the time t; if the position of the particle is smaller than the minimum value Mmin of the position, taking the minimum value as the position of the particle at the time t;

the speed and the position of the particle swarm are updated through 2.7A and 2.7B, and a new particle swarm is generated;

step 2.8, iterative control is carried out on the new particle swarm generated in the step 2.7, so that the optimization has shorter operation time, and a clustering center for primary sorting of radar signals is generated, and the method specifically comprises the following steps:

step 2.8.1, calculating the sorting recognition rate according to the new optimal particle position generated at the moment t; if the sorting identification rate exceeds X, reducing the range of inertial weight;

wherein, the value range of X is 70% to 80%, and the value range of the inertia weight is 0.6 to 1.1 after the range is reduced;

step 2.8.2 judging whether the value exceeds Y according to the identification rate obtained in the step 2.8.1; if the acceleration coefficient exceeds the preset range, reducing the range of the acceleration coefficient;

wherein, the value range of Y is 80% to 90%, and the value range of the acceleration coefficient is 1.0 to 3.0 after the range is reduced;

step 2.8.3, judging whether the value is kept unchanged in T iterations according to the global optimal fitness function value obtained in the step 2.5.2, wherein the value of T does not exceed the maximum iteration times; if the value remains unchanged, the position of the corresponding particle for which the global optimal fitness function value remains unchanged in the T iterations, i.e., the position of the particle that converges to a solution close to the global optimal solution, is recorded. Selecting the position with high recognition rate as the final iteration position of the particle swarm optimization method, namely the final stable primarily sorted clustering center, from the T positions, jumping out of the step 2, and executing the step 3; if the global optimal fitness function value changes, continue to step 2.8.4;

step 2.8.4 updates the iteration number t: t is t + 1; judging whether t is greater than the maximum iteration time Tmax, and if t is greater than Tmax, taking the position of the particle obtained by the last iteration as a clustering center for primary sorting of the radar signal; otherwise, if the Tmax is not greater than the Tmax, jumping to the step 2.4;

step 3, radar signal enhanced data clustering, namely, signal sorting is carried out by adopting a k-means method to finish the enhanced data clustering of radar signals, and the method specifically comprises the following substeps:

step 3.1, initializing each parameter of k-means clustering;

wherein, each parameter includes: cluster center and cluster number; the method specifically comprises the following steps: the cluster centers are initialized to the primarily sorted cluster centers obtained in step 2.8; the cluster number is initialized to the optimal cluster number obtained in step 1.4.2;

step 3.2, calculating the distance between each sample and the initial clustering center, distributing the distance to the nearest neighbor cluster according to the principle of minimum distance, and calculating the distance by adopting the Euclidean distance;

wherein each sample represents radar signals for each of the different parameters;

3.3, calculating the mean value of the samples in each cluster, and taking the mean value as a new cluster center;

step 3.4, repeating the steps 3.2 and 3.3 until the clustering center is not changed any more or the maximum iteration number is reached, and obtaining the final clustering class and the clustering center;

so far, from step 1 to step 3, the radar signal sorting method based on adaptive threshold and iterative control is completed.

Advantageous effects

Compared with the method for sorting the radar signals by combining the particle swarm based on self-adaptive threshold and iterative control and the k-means, the method for sorting the radar signals by combining the particle swarm based on self-adaptive threshold and iterative control has the following beneficial effects:

1. aiming at the problem that the k-means clustering number is uncertain, automatically determining the clustering number by adopting a contour coefficient;

2. aiming at the problems that the k-means initial clustering center is randomized and easily falls into a local optimal solution, a globally optimal initial clustering center is provided for the k-means initial clustering center by adopting a particle swarm method;

3. aiming at the problem of parameter sensitivity of the particle swarm method, the particle swarm method based on the adaptive parameter threshold is provided, so that the particle swarm can be well combined in global optimal search and local optimal search, and the sorting and clustering effects are improved;

4. aiming at the problem that the particle swarm method is long in iteration time for searching the global optimal solution, the particle swarm method with iteration control is provided, so that the iteration times of particles are reduced, and the program running time is shortened;

5. and finally, the k-means method is used again for the clustering center obtained by the particle swarm method, so that the data clustering is enhanced, and the final sorting and clustering effect of the whole method is improved.

6. The particle swarm method has less parameters, does not need operations such as cross variation and the like, has no internal and external circulation, and can control the circulation frequency within a smaller range;

7. compared with other optimization methods, the method has the advantages of fast convergence, less iteration times and higher sorting recognition rate.

Drawings

FIG. 1 is a schematic flow chart of a radar signal sorting method based on adaptive threshold and iterative control PSO and k-means according to the present invention;

FIG. 2 is a space distribution diagram of 6 radar parameters constructed by the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means;

FIG. 3 is a k-means clustering effect graph of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means according to the present invention;

FIG. 4 is a signal sorting result of a particle swarm method of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means according to the present invention;

FIG. 5 is a sorting recognition rate of an improved PSO method of the invention based on an adaptive threshold and iterative control PSO and k-means radar signal sorting method.

Detailed Description

For better illustrating the objects and advantages of the present method, further detailed description of the present application will be provided in conjunction with the accompanying drawings and embodiments.

Example 1

Modern radar signal sorting methods increasingly pay more attention to a plurality of parameters for combined sorting, the more commonly used parameters are, the higher the accuracy of signal sorting results is, and the commonly used signal sorting parameters include pulse arrival angle DOA, pulse width PW and pulse frequency RF. Due to the performance of the radar receiver equipment and the factors of the external environment, the radar pulse parameters have measurement errors, therefore, PDW (pulse duration and width) stream data of the radar can be changed within a certain range, the sorting parameters have the condition of overlapping in different degrees to a certain extent, and the accuracy of radar signal sorting is reduced.

This embodiment illustrates a specific implementation of the method for sorting radar signals based on adaptive threshold and iterative control PSO and k-means according to the present application when sorting radar signals with overlapping parameters, and a flowchart of the implementation of the present application is shown in fig. 1.

The radar characteristic parameters, namely pulse description words PDW, are composed of six parameters, namely pulse arrival angle DOA, pulse width PW, pulse frequency RF, pulse amplitude PA, pulse repetition frequency PRT and pulse arrival time DOA, and in the example, the radar signal sorting is carried out by adopting the three characteristic parameters, namely DOA, PW and RF;

the example uses a mixed sequence of radar signatures with different degrees of overlap of 6 sets of signature parameters generated by a PDW radar signature generator, using the sorting parameters and signal form shown in table 1 below:

TABLE 1 Radar data constructed

During method processing, 500 groups of radar pulse signals are selected from a total generated PDW flow to carry out signal sorting, and parameters such as a final signal sorting average identification rate, a mean square error, clustering evaluation indexes and the like are obtained. The distribution of signal parameters in three dimensions is shown in figure 2.

The radar signal that this application provided is selected separately and is divided into 3 steps, mainly includes: the method comprises three processes of radar pulse PDW stream data preprocessing, radar signal primary sorting and radar signal final enhanced data clustering. Evaluating the number of sample clusters by adopting a contour coefficient, and estimating the effective cluster number of the clusters; performing primary sorting on radar signals by adopting an improved particle swarm method based on adaptive parameter threshold and iterative control to obtain a clustering center; and performing secondary signal sorting on the clustering centers obtained by the particle swarm method in the last step by adopting a k-means method, thereby achieving the effect of enhancing data clustering.

The specific implementation process of radar signal sorting is as follows:

step 1, data preprocessing is carried out on radar pulse streams, and the number of clusters is determined by utilizing contour coefficients;

step 1.1, constructing a radar data set;

the data set comprises data to be sorted and different types of radar signals are generated to form the data to be sorted, and each type of radar signal is a group of PDW stream data; the dimension of the constructed radar data set is D; each radar signal having different parameters represents a data sample;

step 1.2, collecting data to be sorted from the radar data set in the step 1.1; the method specifically comprises the following steps: 500 data samples are selected from the generated mixed radar signal data set to form a radar data sample set to be sorted;

step 1.3, changing different pulse signal data in the data to be sorted into the same scale range, and generating revised data to be sorted; the treatment is as follows:

wherein x represents each data sample; x is the number of_maxRepresents the maximum value of the data sample to be sorted;

step 1.4, evaluating the effective cluster number of the cluster by using the contour coefficient, and recording the effective cluster number as M, wherein the method specifically comprises the following steps: and calculating the contour coefficient on the basis of k-means clustering and obtaining the effective clustering number by using the coefficient. The contour coefficient is a combination of two factors of the intra-class cohesion degree and the inter-class separation degree, and can be used for evaluating the influence of different methods or different operation modes of the methods on the clustering result on the basis of the same original data. The specific calculation method is as follows:

step 1.4.1, calculating a contour coefficient; when in specific implementation, the method comprises the following substeps:

step 1.4.A calculate the average distance from sample i to other samples in the same clusterFrom a_i. Generally, the calculation of the distance uses the euclidean distance. a is_iThe smaller the sample i is, the more the sample i should be clustered to the cluster. A is to_iReferred to as intra-cluster dissimilarity of sample i. A of all samples in cluster c_iThe mean is referred to as cluster dissimilarity for cluster c. The Euclidean distance calculation is as follows:

wherein x and y represent different data samples, and n represents the dimension of the data samples;

step 1.4.B calculate sample i to some other cluster c_jIs the mean euclidean distance b of all samples_ijDefining samples i and clusters c_jThe dissimilarity of (d) is calculated as: b_i＝min{b_i1,b_i2,...,b_ikIn which b is_iThe larger the sample i is, the less the sample i belongs to other clusters.

Step 1.4.C, calculating the contour coefficient of the sample i according to the intra-cluster dissimilarity and the inter-cluster dissimilarity of the sample i:

and further calculating the mean value of the contour coefficients of all the samples, namely the contour coefficient of the clustering result.

Step 1.4.2, a clustering number range is given in advance by adopting an exhaustion method, the value range of the clustering number k is selected to be 3-9, then, in order to avoid falling into a local optimal solution, repeated operation is carried out on each k value for several times, the values of the contour coefficients under the conditions are calculated and compared, the larger the value of the contour coefficient obtained by calculation is, the better the clustering effect of the data sample points is, and the result is shown in the following table 2:

TABLE 2 values of contour coefficients for different numbers of clusters

Number of classification	3	4	5	6	7	8	9
								Coefficient of contour	0.6776	0.7626	0.8185	0.8281	0.8017	0.7726	0.7461

As can be seen from table 2, when the number of clusters is 6, the value of the contour coefficient obtained by clustering is the largest, which indicates that the data is the most compact after clustering. The number of subsequent cluster centers was therefore chosen to be 6.

Step 2, primary sorting of radar signals to generate a clustering center after the primary sorting, which specifically comprises the following steps: searching a global optimal solution, namely a data clustering center after primary sorting, for the data to be sorted revised in the step 1.3 by adopting an improved particle swarm method based on adaptive parameter threshold and iterative control; the method comprises the following substeps:

each parameter includes a population number, a maximum iteration number, a particle position, a maximum movement speed of the particle, and a historical highest fitness function value of the particle, and specifically includes:

the number of particle populations is set to 40, i.e. N is 40; the maximum number of iterations for the particle is set to 100, i.e., Tmax is 100. The position of the particles, i.e. the central point of the initial clustering of the data samples, is initialized to a matrix of dimension M × D × N, M being the optimal number of clusters obtained from the profile coefficients in step 1.4.2, D being determined by the dimension of the data set constructed in step 1.1, where D is 3-dimensional and refers to the angle of arrival DOA, the pulse width PW, and the pulse carrier frequency RF, respectively. Each element in the position matrix is a random number between the maximum value and the minimum value of the data sample obtained in the step 1.3; the particle velocity is a matrix with dimension of M multiplied by D multiplied by N, and each element in the matrix is initialized to be a random number with the value in the range of 0-1; the maximum velocity of the particles was set to 0.8; the historical highest fitness function value of the particle is initialized to infinity;

and 2.2, determining the range of the inertia weight w and the range of the acceleration coefficient c. The range of the two parameters is determined by setting different parameter values and observing the final simulation result. The method comprises the following steps of evaluating by using two indexes of signal sorting identification rate and convergence iteration frequency, wherein the identification rate calculation mode is specifically shown as the following substeps;

step 2.2.1, calculating the distance from the sample point to the clustering center obtained by each method, and finding the clustering center closest to the sample, wherein the center represents the clustering division center obtained by the method;

step 2.2.2 calculate the distance from the clustering center obtained in the last step 2.2.1 to the reference real clustering center, and find the corresponding real class center closest to the center;

and 2.2.3, comparing whether the clustering center obtained by the data sample through the method is the same as the corresponding real clustering center. If the data samples are of the same type, the data samples are correctly clustered; if the data samples are not of the same type, indicating that the data samples are clustered wrongly;

step 2.2.4, counting the number of correct clusters of the whole radar signal sample, calculating the proportion of the number in the whole data sample, and taking the result obtained by calculation as the radar signal sorting identification rate;

tables 3 and 4 below show the effect of inertial weight w and acceleration coefficient c on the particle swarm method:

(1) the influence of different values of the inertial weight w on the particle swarm optimization method is shown in the following table 3:

TABLE 3 influence of inertial weights w on particle swarm method

Inertial weight w	0.4	0.6	0.8	1.0	1.2	1.4
							Average sorting recognition rate	84.44％	87.52％	89.06％	84.74％	72.54％	60.36％
Mean convergence iteration numberNumber of	96	81	35	22	18	11

(2) The influence of different values of the acceleration coefficient c on the particle swarm optimization method is shown in the following table 4:

TABLE 4 influence of the acceleration coefficient c on the particle swarm method

Coefficient of acceleration c	1.0	1.5	2.0	2.5	3.0	3.5	4.0
								Average sorting recognition rate	89.58％	86.06％	89.06％	83.44％	85.58％	80.12％	77.34％
Average number of converged iterations	81	55	35	24	25	20	17

Step 2.2.5 determines the range of the inertial weight coefficient w and the acceleration coefficient c. According to the above tables 3 and 4, under the condition that the program running time is relatively short, the iteration times of the particle swarm are as few as possible, and the recognition rate is high enough, the dynamic range of the parameters is set as follows: the initial inertial weight w ranges from w_ini＝1.0，w_end0.6; the range of the initial acceleration coefficient c is c_ini＝3.0，c_end＝1.5；

Step 2.3, setting a cycle variable t to be 1, and taking the initial values of the position and the speed of the particle swarm as the position and the speed of the particle swarm at the time t-1, namely the time 0;

step 2.4, calculating a fitness function value of each particle at the time t, specifically: and calculating the Euclidean distance between the cluster center and the particle position obtained by the particle swarm method at the t-1 moment, namely the cluster center and the corresponding reference cluster center, namely the mean square error value between the cluster center and the particle position, namely the fitness function value. The smaller the fitness function value is, the smaller the error of the clustering center obtained by the function from the real reference clustering center is, and the better the performance of the particle is; when t is 1, the position of the particle at the time t-1, namely at the time 0, is the initialized particle position;

step 2.5, calculating an optimal value pbest in the particles at the current time t and a global optimal value gbest of the particle swarm, specifically;

step 2.5.1, calculating the optimal position pbest in the particle at the current time t, namely selecting the position which enables the fitness function value of the particle to be the minimum as the optimal position; for each particle, according to the fitness function value of the t moment obtained by calculation in the last step 2.4, the fitness function value of the optimal position of the particle at the t-1 moment is compared with the value, if the fitness function value is better, namely the fitness function value is smaller, the optimal position of the particle is updated, if the fitness function value is not better than before, the optimal position of the particle is not updated, and the optimal position of the particle at the last moment, namely the t-1 moment is kept; when t is 1, the historical highest fitness function value of the particle at the time t-1, namely the time 0, is the initialized historical highest fitness function value;

step 2.5.2, calculating a global optimal position gbest in the particle swarm, namely comparing the fitness function value corresponding to the optimal position of each particle obtained in the step 2.5.1, selecting the optimal position in the whole particle swarm with the minimum fitness function value, and marking the found optimal position with the minimum fitness function value as the global optimal fitness function value of the particle swarm;

and 2.6, updating the inertia weight w and the acceleration coefficient c of the particle swarm. The inertia weight is used for controlling the influence of the previous speed on the current speed, the larger w can enhance the global searching capability of the particle swarm optimization method, and the smaller w can enhance the local searching capability. The acceleration coefficient c is used for characterizing the influence of social experience and cognitive experience of the particles on the positions of the particles. A low value indicates that the particle may wander outside the target region before being pulled back, while a high value causes the particle to suddenly rush toward or over the target region; the updating mode specifically comprises the following substeps:

step 2.6.1, updating the inertia coefficient w of the particle swarm; as can be seen from table 3 above, as the weight coefficient is gradually increased, the sorting recognition rate is gradually decreased, and the average number of iterations for achieving convergence is gradually decreased. As the weighting factor is gradually decreased, the sorting recognition rate is also decreased, but the average number of iterations to reach convergence is gradually increased. According to the change, the updating method for obtaining the inertia coefficient is as follows, namely the updating method is a linear increasing updating method:

w(t)＝(w_ini-w_end)*t/t_max+w_end (4)

wherein, w_iniIs the inertial weight value, w, at the first iteration_endTo reach the weight value at the last iteration, i.e. the range of inertial weights determined in step 2.2.5, t is the number of iterations, t_maxIs the maximum iteration number, which is the value of the iteration number initialized in step 2.1;

step 2.6.2, updating the acceleration coefficient c of the particle swarm; as can be seen from table 4 above, the acceleration coefficient has a smaller influence on the sorting recognition rate than the weight coefficient, but has a larger influence on the convergence rate. According to the change, an updating mode of the acceleration coefficient is obtained, and the updating mode of linear decrement is shown as (5):

c₁(t)＝(c_1ini-c_1end)*(t_max-t)/t_max+c_1end

c₂(t)＝(c_2ini-c_2end)*(t_max-t)/t_max+c_2end (5)

wherein, c_1ini c_2iniIs the weight value at the first iteration, c_1end c_2endThe weight value when the last iteration number is reached, namely the range of the acceleration coefficient determined in the step 2.2.5;

where t is the number of iterations, t_maxIs the maximum iteration number, which is the value of the iteration number initialized in step 2.1;

step 2.7, updating the position and the speed of the particle swarm to generate a new particle swarm; the updating method comprises the following specific steps:

step 2.7.1 speed of updating particle swarm, the updating mode is as follows:

V_i(k+1)＝w*V_i(k)+c₁*rand*(X_pbest,i-X_i)+c₂*rand*(X_gbest,i-X_i) (6)

wherein, X_pbestAnd X_gbestRespectively representing the better value of the k generation individual and the best value in the solution group; w and c are the values of the inertial weight and acceleration coefficient obtained in step 2.6 above; rand represents a random number in the range of 0-1; the first part of the equation is a memory term, which represents the influence of the last speed magnitude and direction; the second part is a self-recognition item, which is a vector pointing to the best position point pbest of the particle from the particle position point at the time t-1 and represents that the motion of the particle comes from self experience; the third part is a group recognition item, which is a vector calculated by pointing the best position point gbest of the group by the particle position point at the time of t-1 and reflects the cooperative cooperation and knowledge sharing among the particles.

If the updating speed of the particles is greater than the maximum speed of the particles obtained in the step 2.1, the updating speed at the current time t is taken as a maximum speed value; if the updating speed of the particles is less than the negative value of the maximum speed of the particles obtained in the step 2.1, taking the updating speed at the current time t as the negative value of the maximum speed value;

step 2.7.2, updating the positions of the particle swarm, wherein the updating mode is as shown in (7):

X_i(k+1)＝X_i(k)+V_i(k+1) (7)

the formula shows that the update of the current t-time position of the particle is related to the position of the last time, namely t-1 time and the speed of the current t time;

step 2.8, iterative control is carried out according to the new particles generated in step 2.7, so that the optimization has shorter running time, and a clustering center for primary sorting of radar signals is generated, and the specific steps are as follows:

step 2.8.1, calculating the sorting recognition rate according to the new optimal particle position generated at the moment t; if the sorting recognition rate exceeds 75%, the range of the inertial weight is reduced. As is found from table 3 above, the range of inertial weight becomes 0.7 to 0.9;

step 2.8.2 judges whether the value exceeds 85% based on the recognition rate obtained in the previous step 2.8.1. If so, the range of the acceleration coefficient is again reduced on the basis of the reduction of the inertial weight. As is found from table 4 above, the range of the acceleration coefficient becomes 1.5 to 2.5;

step 2.8.3, according to the global optimal fitness function value obtained in step 2.5.2, judging whether the value is kept unchanged in 10 iterations; if the value remains unchanged, the position of the corresponding particle for which the global optimal fitness function value remains unchanged in 10 iterations, i.e. the position of the particle that converges to a solution close to the global optimal solution, is recorded. Selecting the position with high recognition rate as the final iteration position of the particle swarm optimization method, namely the final stable primarily sorted clustering center, from the 10 positions, jumping out of the step 2, and executing the step 3; if the value changes, the following steps are continued;

step 2.8.4 updates the iteration number t: t is t + 1; and judging whether t is greater than the maximum iteration time Tmax. If the position of the particle is larger than Tmax, the position of the particle obtained by the last iteration is used as a clustering center for primary sorting of the radar signal; if not, skipping to step 2.4;

3, carrying out final signal sorting work by using a k-means method to finish the final data clustering function; the method comprises the following specific steps:

step 3.1, initializing each parameter of the k-means method;

wherein, each parameter includes: cluster center and cluster number; the method specifically comprises the following steps: taking the primarily sorted clustering center obtained in the last step 2.8 as an initial clustering center of a k-means method; taking the optimal clustering number obtained in the step 1.4.2 as the clustering number of the k-means method;

step 3.4, repeating the steps 3.2 and 3.3 until the clustering center is not changed or the maximum iteration number is reached, and obtaining a final clustering result and a clustering center;

after the radar signal sorting method is completed, the method is implemented in the following steps: evaluating the clustering result obtained in the step 3.4 based on the evaluation index;

the evaluation indexes comprise identification rate, mean square error and external evaluation indexes, and the external evaluation indexes comprise JC, FMI and RI parameters.

Wherein the external evaluation index comprises JC, FMI and RI parameters; the specific calculation steps are as follows:

step 4.1, calculating JC coefficients; for dataset D ═ x₁,x₂,...,x_mSuppose that clusters obtained by clustering are divided into C ═ C₁,C₂,...,C_KDivision of clusters given by the reference model into C^*＝{C^* ₁,C^* ₂,...,C^* _mLet λ, λ respectively^*Respectively represent and C, C^*Corresponding cluster mark vectors are defined by considering pairwise matching of samples

Defining the JC coefficient is obtained by the following equation:

jaccard Coefficient (JC):

step 4.2, calculating an FM index; the a, b, c, d parameters found in step 4.1 can be obtained from the following equations:

FM index (FMI):

step 4.3, calculating RI index; the a, b, c, d parameters found in step 4.1 can be obtained from the following equations:

rand Index (RI):

step 5 simulates two other typical optimization methods for comparison, namely a genetic method and a simulated annealing method, in order to illustrate the effectiveness of the proposed method. And (4) similarly calculating various cluster evaluation indexes related to the step (4) by using the preprocessed data for the two methods. For the genetic method, under the condition of ensuring a certain accuracy, the parameters are set as follows: maximum number of iterations t _max100; the population number particles is 40; the mutation probability _ rate is 0.1, and the crossover probability is 0.5. For the simulated annealing method, the set parameters are as follows: initial temperature T ═ 100; the ending temperature T _ min is 0.0001; the iteration number L at each temperature is 20; the annealing coefficient a is 0.95.

All steps of the radar signal sorting method based on adaptive threshold and iterative control are completed.

Table 5 below compares the modified PSO method + k-means method with the sort recognition rates obtained using the two methods alone:

TABLE 5 comparison of the Effect of the improved particle swarm method in combination with the k-means method

As can be seen from table 5 above, the improved particle swarm optimization method and the k-means method are combined to greatly improve the sorting recognition rate compared with the improved particle swarm optimization method, the improved particle swarm optimization method and the k-means method, which are used alone, and therefore, the method is feasible for signal sorting.

The following table 6 shows the sorting results of different optimization methods and k-means;

TABLE 6 sorting results combining different optimization methods with k-means

As can be seen from table 6 above, the recognition rate of the particle swarm optimization method is the highest when the optimization methods are used alone, which indicates that the method is advantageous and effective compared to other optimization methods. The results of the several optimization methods and the combination of k-means can all achieve better results. From the aspect of the discrimination rate, although the discrimination rate obtained by the simulated annealing method is the highest, the time consumption is the longest due to the existence of the internal and external circulation. Although the particle swarm method is not optimal in recognition rate, when the recognition rate is not very different, the time consumption is the shortest and the convergence is the fastest. Therefore, under the condition of comprehensively considering both the recognition rate and the simulation time, the improved PSO method is selected as the method for sorting the radar signals. Meanwhile, compared with the method of singly using k-means clustering, the method improves the clustering by more than 10 percent, improves the recognition rate to a certain extent compared with the method of singly using PSO, and greatly reduces the mean square error.

Tables 7 and 8 below are clustering sorting evaluation indexes under various methods and sorting evaluation indexes combining different methods with k-means:

TABLE 7 clustering sorting evaluation indexes under various methods

TABLE 8 evaluation index of sorting combining different methods with k-means

As can be seen from tables 7 and 8 above, the index value obtained by the standard PSO method is larger than that obtained by the k-means method, indicating that the PSO method can make the data clustering more compact. The PSO method has good clustering effect when the optimization methods are used independently. As can be seen from Table 8 above, after the optimization method + k-means clustering is used, the index parameters are further improved, so that the intra-class distance of the data is smaller and the inter-class distance is larger. This also illustrates that the final use of the k-means method serves as a cluster enhancement for the data. In conclusion, the improved PSO method and the k-means method are combined to obtain a better clustering effect.

FIG. 2 is a space distribution diagram of 6 radar parameters constructed by the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means; it is evident from the figure that for radar pulse signals, there is a different degree of overlap between the three characteristic parameters, angle of arrival, pulse width and carrier frequency. By using some traditional clustering methods, good sorting results cannot be obtained;

FIG. 3 is a k-means clustering effect graph of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means; it can be seen from the figure that the k-means method has instability, which is caused by the randomness of the initial cluster center selection, so that the method has great influence on each time different initial values are selected;

fig. 4a is an optimal fitness function value of the particle swarm iteration of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means in the present application (for example, w ═ 0.8, c ═ 2); fig. 4b shows the signal clustering sorting recognition rate and the mean square error (w is 0.8, c is 2) of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means according to the present application; as can be seen from fig. 4a, the fitness function value of the particle is gradually decreased in each iteration process, which indicates that the clustering center obtained by the method is gradually close to the real clustering center, and the mean square error of the clustering center is gradually reduced; as can be seen from fig. 4b, the particles are finally iterated to converge in the continuous optimization, so that the signal sorting recognition rate reaches a higher value, and the square mean square error is not changed;

FIG. 5a is a sorting recognition rate of an improved PSO method of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means; FIG. 5b is a sorting recognition rate of an improved PSO + k-means method of the radar signal sorting method based on adaptive threshold and iterative control PSO and k-means; as can be seen from FIG. 5a, compared with k-means clustering, the particle swarm method has higher stability, and the recognition rate can be basically kept above 90%; as can be seen from FIG. 5b, for the method combining particle swarm and k-means, the re-clustering of the k-means method can make the data sample clustering effect better, can improve the final recognition rate, and plays a role in enhancing data clustering.

While the foregoing is directed to the preferred embodiment of the present invention, it is not intended that the invention be limited to the embodiment and the drawings disclosed herein. It is intended that all equivalents and modifications which come within the spirit of the disclosure be protected by the present invention without departing from the spirit of the disclosure.

Claims

1. The radar signal sorting method based on the self-adaptive threshold and the iterative control is characterized by comprising the following steps of: the method comprises the following steps: the method comprises the following steps of radar pulse PDW stream data preprocessing, radar signal primary sorting and radar signal enhanced data clustering:

step 1, preprocessing radar pulse PDW flow data to generate effective cluster number, specifically comprising constructing a radar data set, collecting and revising data to be sorted and evaluating the effective cluster number of clustering based on the profile coefficient;

step 1, specifically comprising the following substeps:

step 1.1, constructing a radar data set;

step 1.4, calculating a contour coefficient, and evaluating the effective cluster number of the cluster based on the contour coefficient;

step 2, comprising the following substeps:

recording the population number as N; the maximum iteration number is recorded as Tmax; the positions of the particles are position matrixes of M multiplied by D multiplied by N dimensions, and elements in the position matrixes are random numbers between Mmin and Mmax; a velocity matrix with a particle velocity of dimension M × D × N;

in step 2.2, the inertial weight is marked as w; the acceleration coefficient is recorded as c;

step 2.8.3, judging whether the value is kept unchanged in T iterations according to the global optimal fitness function value obtained in the step 2.5.2, wherein the value of T does not exceed the maximum iteration times; if the value is kept unchanged, recording the position of the corresponding particle with the global optimal fitness function value kept unchanged in the T iterations, namely the position of the particle converging to the global optimal solution; selecting the position with high recognition rate as the final iteration position of the particle swarm optimization method, namely the final stable primarily sorted clustering center, from the T positions, jumping out of the step 2, and executing the step 3; if the global optimal fitness function value changes, continue to step 2.8.4;

step 3.1, initializing each parameter of k-means clustering;

wherein, each parameter includes: cluster center and cluster number;

2. The adaptive threshold and iterative control-based radar signal sorting method of claim 1, wherein: in step 1.1, the dimension of the constructed radar data set is D, the radar data set comprises data to be sorted, which are formed by generating different types of radar signals, and each type of radar signal is a set of PDW stream data.

3. The adaptive threshold and iterative control-based radar signal sorting method of claim 2, wherein: in step 1.3, the maximum value of the data to be sorted is recorded as Mmax, and the minimum value of the data to be sorted is recorded as Mmin.

4. The adaptive threshold and iterative control-based radar signal sorting method of claim 3, wherein: step 1.4, specifically:

step 1.4.1, calculating a contour coefficient;

and recording the optimal clustering number as M.

5. The adaptive threshold and iterative control-based radar signal sorting method of claim 4, wherein: in step 2.1, the value range of the population number N is 10 to 50; the maximum iteration number is in the range of 50 to 200; each element in the speed matrix is a random number with the value in the range of 0-1, and the maximum value in the speed matrix is between 0.5 and 1.0; the historical highest fitness function value for the particle is initialized to infinity.

6. The adaptive threshold and iterative control-based radar signal sorting method of claim 5, wherein:

in the step 2.2, the value range of the inertia weight w is between 0.5 and 1.2; the value range of the acceleration coefficient c is between 0 and 4.

7. The adaptive threshold and iterative control-based radar signal sorting method of claim 6, wherein: in step 2.8.1, the value range of X is 70% to 80%, and the value range of the inertial weight is 0.6 to 1.1 after the range is reduced.

8. The adaptive threshold and iterative control-based radar signal sorting method of claim 7, wherein: in step 2.8.2, the value range of Y is 80% to 90%, and the value range of the acceleration coefficient is 1.0 to 3.0 after the range is reduced.

9. The adaptive threshold and iterative control-based radar signal sorting method of claim 8, wherein: step 3.1, specifically: the cluster centers are initialized to the primarily sorted cluster centers obtained in step 2.8; the cluster number is initialized to the optimal cluster number obtained from step 1.4.2.

10. The adaptive threshold and iterative control-based radar signal sorting method of claim 9, wherein: in step 3.2, each sample represents the radar signal for each of the different parameters.