CN113866735A

CN113866735A - Radar signal sorting method based on dynamic correction chaotic particle swarm optimization

Info

Publication number: CN113866735A
Application number: CN202110682277.7A
Authority: CN
Inventors: 傅雄军; 王晓妍; 谢民; 马志峰; 卢继华; 姜嘉环
Original assignee: Beijing Institute of Technology BIT
Current assignee: Beijing Institute of Technology BIT
Priority date: 2021-06-20
Filing date: 2021-06-20
Publication date: 2021-12-31

Abstract

The invention relates to a radar signal sorting method based on dynamic modified chaotic particle swarm optimization, and belongs to the technical field of population evolution and signal classification. Aiming at the problems of high pulse stream density and serious overlapping degree of characteristic parameters of radiation source signals in a complex electromagnetic environment, the radar signal sorting method based on dynamic correction chaotic particle swarm optimization is adopted, and the defects that the traditional clustering sorting algorithm is difficult to correctly sort and the particle swarm optimization is insufficient are overcome. Chaos search is adopted to increase the diversity of later iteration of the population; the updating of the particles is changed in real time according to the state of the population by adopting the self-adaptive adjustment parameters; and the new fitness function is used and the positions of the particles are dynamically corrected, so that the population optimization is more accurate. Compared with other optimization methods, the method has the advantages of being high in convergence rate, stability and robustness, good in sorting effect and capable of being better adapted to complex electromagnetic environments under the condition of several common and new sorting indexes.

Description

Radar signal sorting method based on dynamic correction chaotic particle swarm optimization

Technical Field

The invention relates to a radar signal sorting method based on dynamic modified chaotic particle swarm optimization, and belongs to the technical field of population evolution and signal classification.

Background

With the increasing complexity of electromagnetic environments, dense and diverse radiation source signals enter a digital scout receiver and are interleaved into a complex pulse stream sequence. And sorting the pulse streams according to the information such as the characteristic parameters, arrival time and the like of the intercepted pulses, accurately dividing the signals of the same radiation source, and identifying the radar model according to the characteristic parameters of different radiation sources. And obtaining the type, the attribute and the threat degree of each radar according to the identification result. From the above analysis, it can be seen that the source sorting is a key link for radar reconnaissance signal processing, and directly affects the performance of radar reconnaissance equipment, which is related to subsequent combat decisions. Therefore, efficient and accurate signal sorting methods are of great importance.

Signal sorting techniques are mainly divided into two aspects based on Pulse Repetition Interval (PRI) analysis and feature clustering. Modern acquisition systems often combine these two methods to achieve signal sorting. However, due to the wide application of PRI agility and low interception, the TOA of the intercepted signal is greatly jittered and lost, thereby destroying the statistical characteristics of the TOA difference. These reasons greatly increase the difficulty of sorting based on PRI analysis. Therefore, a clustering method based on radar characteristic parameters is developed at the same time and becomes an indispensable task for the de-interleaving process of the auxiliary pulse signals.

Compared with the traditional clustering sorting method, the heuristic method is a group method, and improves the sorting performance by searching global optimum. However, although the PSO method provides the possibility of global search, convergence to global optimum cannot be guaranteed, and there are also disadvantages such as premature convergence and insufficient optimization capability. The main current trend to solve these problems is to increase the diversity of the population or to merge other methods, etc. For example, some scholars propose a method based on dynamic particle swarm optimization and K-means, which enhances the performance of the method by dynamically adjusting inertial weight and acceleration coefficient. Some scholars improve the cluster centroid by changing the fitness function. However, they are not completely adaptive, and the clustering result cannot well meet the requirements of the radar system on signal sorting accuracy and instantaneity.

In order to accurately separate various radiation source signals from dense, complex and changeable intercepted pulse streams, the application provides a Radar Signal Sorting method based on Dynamic Modified Chaos Particle Swarm (DMCPSO), so that the defects of difficult Sorting and PSO (Particle Swarm Optimization) Optimization of the traditional clustering Sorting method are overcome, and the Sorting performance is improved. Simulation results show that the particle swarm optimization method has great advantages compared with other improved particle swarm optimization methods under several common and newer sorting indexes, and has better sorting effect in convergence speed, stability and robustness.

Disclosure of Invention

The invention aims to accurately separate various radiation source signals from a dense, complex and variable intercepted pulse stream, improve the technical situations that the traditional clustering separation method is difficult to correctly classify, particle swarm optimization is easy to early-mature and converge, and the optimizing capability is insufficient, and provide a signal separation method based on a dynamic correction chaotic particle swarm radar.

In order to achieve the purpose, the invention adopts the following technical scheme.

The radar signal sorting method comprises the following steps:

step 1, preprocessing radar pulse PDW flow data, constructing a radar data set and generating revised data to be sorted;

preprocessing radar pulse PDW flow data, wherein the preprocessing comprises constructing a radar data set, collecting data to be sorted and revising the data to be sorted;

step 1 specifically comprises the following substeps:

step 1.1, constructing a radar data set;

wherein 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, the number of the types is recorded as M, 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 the step 1.1, and recording the total number of the data to be sorted as S;

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 2, radar signal sorting and generation of a radar signal sorting clustering center, namely a sorted result, specifically comprising: searching a global optimal solution, namely a sorted data clustering center, for the revised to-be-sorted data generated in the step 1.3 by adopting a chaotic particle swarm optimization method based on dynamic correction; the method comprises the following substeps:

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

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

wherein the population number is recorded as N, and the value range of N is 10 to 20; the value range of the maximum iteration times is 50 to 200 and is marked as Tmax; the particle velocity is a velocity matrix with M multiplied by D multiplied by N dimensions, each element in the velocity matrix is a random number with the value ranging from 0 to 1, and the maximum value in the velocity matrix is between 0.5 and 1.0; the historical highest fitness function value of the particle is initialized to 0;

step 2.2, initializing the position of the M multiplied by D-dimensional particles by using a Maxmin distance principle to obtain an initial clustering center of the population, and comprising the following substeps:

step 2.2A, selecting a sample from any constructed radar data set as a first clustering center;

step 2.2B, selecting a sample farthest from the first clustering center as a second clustering center;

2.2C, selecting a point with the maximum nearest distance from the first clustering center and the second clustering center as a central point of a third initial cluster, and repeating the steps until M initial cluster central points are selected and are used as initial clustering centers of the population;

step 2.3, an iteration loop variable t is set, and an initial loop variable t is set to be 1;

wherein, the iteration loop variable t represents t moment;

step 2.4, calculating the clustering division of the data set at the time t and calculating the particle positions, specifically: dividing all samples in the data set according to the Euclidean distance minimum principle with the particle position, and calculating an average value of the samples in each class of clusters according to the divided result to serve as a new particle position;

step 2.5, calculating the fitness function value of each particle, specifically comprising the following steps:

step 2.5A, calculating the sum of the distances between all the data samples contained in only one cluster and the cluster centroids thereof, and recording the sum as E1;

step 2.5B, calculating the sum of the intra-cluster distances between the samples and the cluster centroids thereof, and recording the sum as E;

step 2.5C, calculating the maximum value of the distance between the samples, and recording the maximum value as D;

step 2.5D, dividing E1 by M, E and D to obtain a fitness function value at the time t;

step 2.6, for each particle, obtaining a fitness function value at the time t according to the step 2.5, comparing the fitness function value with the fitness function value of the optimal position at the time t-1 which the fitness function value has undergone, if the fitness function value is better, namely the fitness function value is larger, updating the optimal position of the particle, and correcting the position of the particle according to the gravity index to obtain the corrected position of the particle; if the fitness function value at the time t is not greater than the fitness function value at the time t-1, the optimal position of the particle is not updated, and the optimal position which is passed by the last time, namely the time t-1, is continuously maintained;

wherein, the particle position is corrected according to the gravity index, and the method comprises the following substeps:

step 2.6A, calculating the sum of the distances between all sample points, and recording the sum as d;

step 2.6B, calculating the neighborhood radius of the sample, and recording the neighborhood radius as r, specifically as follows: dividing the sum of the distances d obtained in the step 2.6A by the power a of S and the power a of S-1;

wherein, the value range of a is 0 to 1;

step 2.6C, calculating a set of the samples and neighbors which are away from the samples by r, and obtaining the gravity index of each in-class sample;

step 2.6D, comparing the size of the gravity index value of the sample including the particle position in the class, and selecting the particle position with the largest gravity index as the corrected particle position;

step 2.6E, performing class separation inside each cluster, specifically: if the distance between the two samples is greater than a certain threshold value, the distance is marked as g, and the gravity index of the two samples is positioned in the front h, the two samples are taken as a new clustering center, namely a new particle position;

wherein, the value range of g is 2 to 4 times of r, and the value range of h is 20 to 40 percent;

step 2.6F merges all the cluster centers found in step 2.6E, specifically: if the two nearest clustering centers are merged, the new clustering center after merging is the mean value of the two clustering centers; repeating the step 2.6F until M clustering centers are reached, wherein the M clustering centers are the corrected particle positions;

step 2.7, dividing all samples in the data set according to the Euclidean distance minimum principle with the obtained corrected particle position;

step 2.8 executes step 2.5, recalculates the fitness function value of each corrected particle;

step 2.9, calculating the global optimum value gbest of the particle swarm at the current time t, specifically: selecting the global optimal particle with the maximum fitness function value, wherein the corresponding position of the global optimal particle is the global optimal value gbest;

step 2.10, randomly disturbing the positions of the particles by using Tent chaotic search to generate a chaotic sequence, and generating new particles according to the generated chaotic sequence, which specifically comprises the following substeps:

step 2.10A, normalizing the global optimal value gbest obtained in the step 2.9 from an optimized variable value interval [ Mmin, Mmax ] to a value interval [0,1] of a chaotic variable;

step 2.10B, performing J-times chaotic disturbance on the global optimal value gbest to generate a chaotic sequence, which specifically comprises the following steps: when the value of the chaotic variable is less than y, the generated chaotic sequence is twice of the original chaotic variable, and when the value of the chaotic variable is more than y, the generated chaotic sequence is obtained by subtracting twice of the original chaotic variable from 2;

step 2.10C, generating new particles according to the generated chaotic sequence, wherein the particles are equal to a global optimal value gbest plus k times of a chaotic variable, and the value of k is 1 to 10 percent;

step 2.11, calculating a fitness function value of the new particles generated according to the chaotic sequence according to the step 2.5, and selecting the chaotic sequence with the maximum fitness function value to replace the particle individuals which are optimal and not global optimal in the current particle swarm;

step 2.12, updating the acceleration coefficients c1 and c2 of the particle swarm, specifically: calculating the ratio between the current iteration t moment and the maximum iteration time Tmax, solving an inverse index of the ratio, performing double operation to obtain an acceleration coefficient c1, and solving an index of the ratio to obtain an acceleration coefficient c 2;

step 2.13, updating the inertia weight w of the particle swarm, specifically comprising: calculating the ratio of the fitness function value of the particles at the current t moment to the global optimal fitness function value, and calculating the inverse index of the ratio to obtain an inertia weight w;

step 2.14, updating the speed and the position of the particle swarm to generate a new particle swarm; the updating method comprises the following specific steps:

step 2.14A, updating the particle swarm speed, specifically: 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 w by the speed of the particles at the time 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 a global optimal value gbest by an acceleration coefficient and a random number in a range from 0 to 1, wherein the particle position point points to the best position point of the group at the moment of t-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 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 2.14B, updating the positions of the particle swarms, wherein the updating mode specifically comprises the following steps: adding the position of the particle at the t-1 moment and the speed of the particle at the current t moment to obtain the position of the particle at the t moment; judging whether the position of the particle at the time t is larger than the maximum value Mmax of the position, if so, taking the maximum value Mmax as the position of the particle at the time t; otherwise, if the position of the particle at the time t is smaller than the minimum value Mmin of the position, taking the minimum value Mmin as the position of the particle at the time t;

step 2.15, calculating the variance of the population fitness function value;

step 2.16, judging whether an iteration termination condition is reached, specifically: comparing whether the variance is smaller than a certain threshold value v or whether the comparison time t is larger than the maximum iteration time Tmax;

wherein, the value range of the threshold value v is 0.2 to 0.8 times of the fitness variance of the population with the time t being 1;

if the two conditions meet one or both of the two conditions, the position of the particle at the time t is used as a clustering center for sorting the radar signals; if neither is satisfied, update t: t is t +1, and the step 2.4 is skipped;

step 3, evaluating the clustering result obtained in the step 2 based on the evaluation index;

the evaluation indexes comprise clustering quality, adjusted lander indexes, normalized mutual information, centroid indexes, Davison bauxid indexes and outline indexes;

so far, from step 1 to step 3, the radar signal sorting method based on the dynamic modified chaotic particle swarm optimization is completed.

Advantageous effects

Compared with the existing large signal sorting method, the chaotic particle swarm optimization radar signal sorting method based on dynamic correction has the following beneficial effects:

1. aiming at the problem that population diversity is lost and easily falls into local optimum in the iterative later stage of the particle swarm optimization method, Tent chaotic search is added; and aiming at the phenomena of small period and unstable period points which are not beneficial to the optimization problem in Tent, the chaotic mapping is improved;

2. aiming at the problem that the particle swarm method cannot be adaptively adjusted by adopting fixed parameters, the method adopts adaptive adjustment parameters (inertia weight and acceleration coefficient) related to a fitness value to sense the change of a swarm in real time, so that the global search capability and the local search capability of particles are better combined;

3. aiming at the problems that the fitness function is single in form, less in contained information and greatly influenced by discrete points and characteristic distribution caused by the fact that parameters of the particle swarm method only use internal distances, a new fitness function is provided, and the sorting accuracy is improved;

4. aiming at the problems that the particle swarm method is sensitive to discrete points and cannot accurately sort, the positions of particles are adjusted through cluster analysis, and then the gravity index is introduced to carry out final position correction on the particles meeting the conditions, so that the particles are closer to a real cluster center;

5. the method aims at 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; therefore, 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 dynamically modified chaotic particle swarm according to the present invention;

FIG. 2 is a radar parameter space distribution diagram of the radar signal sorting method based on the dynamic modified chaotic particle swarm;

FIG. 3 is a signal sorting result of the radar signal sorting method based on the dynamic modified chaotic particle swarm in the invention;

fig. 4 is a adaptability variance curve diagram of different methods of the radar signal sorting method based on the dynamic modified chaotic particle swarm.

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.

The embodiment illustrates specific implementation of the method for sorting radar signals based on the chaos particle swarm optimization with dynamic correction when radar signals with overlapped parameters are sorted, and an implementation flow chart of the method 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 8 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

When the method is used for processing, 1000 groups of radar pulse signals are selected from the total generated PDW flow to carry out signal sorting, and parameters such as final signal sorting running time and various sorting evaluation indexes 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 2 steps, mainly includes: the method comprises two processes of radar pulse PDW flow data preprocessing and radar signal sorting based on dynamically corrected chaotic particle swarms.

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

step 1, preprocessing radar pulse flow;

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 dimensionality of the constructed radar data set is D, and the class number is M; 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: selecting S data samples from the generated mixed radar signal data set to form a radar data sample set to be sorted, wherein the value of S is 1000;

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; xmax represents the maximum value of the data sample to be sorted, and xmin represents the minimum value of the data sample to be sorted;

step 2, radar signal sorting and generating sorted results, which specifically comprise: searching a global optimal solution, namely a sorted data clustering center, for the revised to-be-sorted data generated in the step 1.3 by adopting a chaotic particle swarm optimization method based on dynamic correction; the method comprises the following substeps:

step 2.1 initialize each parameter of the particle swarm optimization method. Each parameter includes a population number, a maximum iteration number, a particle speed, 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 10, i.e. N is 10; the maximum number of iterations for the particle is set to 50, i.e., Tmax is 50. The velocity of the particles is initialized to a matrix of dimensions M x D x N, M being determined by the class number of the dataset constructed in step 1.1 and D being determined by the dimension of the dataset constructed in step 1.1, where D is 3-dimensional and refers to the angle of arrival DOA, the pulse width PW, the pulse carrier frequency RF, respectively. Each element in the velocity matrix is a random number between the maximum value and the minimum value of the data sample obtained in the step 1.3; the maximum velocity of the particles was set to 0.8; the historical highest fitness function value of the particle is initialized to 0;

step 2.2, initializing the particle position by using a Maxmin distance principle, wherein the particle position is in dimension of M multiplied by D and is divided into the following substeps;

step 2.2B, selecting a sample farthest from the first clustering center as a second clustering center, wherein the distance adopts an Euclidean distance, and the calculation mode is as follows:

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

2.2C, selecting a point with the maximum closest distance from the first clustering center to the second clustering center as a central point of a third initial cluster, and repeating the steps until M initial cluster central points are selected and are used as initial clustering centers of the population;

step 2.3 sets an iteration loop variable t, i.e. time t. Setting an initial circulation variable as t-1;

step 2.5, calculating the fitness function value IPBM of each particle, which specifically comprises the following steps: the sum of the distances between all data samples and their cluster centroids is calculated as E1. And calculating the sum of the intra-cluster distances between the sample and the cluster centroid, and recording as E. The maximum value of the distance between the samples is calculated and is denoted as D. The fitness function value IPBM computing method is as follows:

step 2.6, for each particle, obtaining a fitness function value at the time t according to the step 2.5, comparing the fitness function value with the fitness function value of the optimal position at the time t-1, which is passed by the fitness function value, if the fitness function value is better, namely the fitness function value is larger, updating the position of the particle, and correcting the position according to the gravity index; if the fitness function value is not better than before, the optimal position of the particle is not updated, and the optimal position that was experienced at the last time, i.e., at time t-1, is continuously maintained. Wherein, the correction of the particle position according to the barycentric index is divided into the following sub-steps:

step 2.6B, calculating the neighborhood radius of the sample, and recording the neighborhood radius as r, specifically as follows: dividing the distance sum d obtained in the step 2.6A by the power a of S and the power a of S-1, wherein the value of a is 0.3, and the calculation mode is shown in a formula (4);

step 2.6C, calculating a set of the samples and neighbors which are r away from the samples to obtain the gravity index GCI of each in-class sample, wherein the calculation mode is as follows:

step 2.6D, comparing the size of the gravity center index of the sample including the particle position in the class, and selecting the particle position with the largest gravity center index as the corrected particle position;

step 2.6E, performing class separation inside each cluster, specifically: if the distance between the two samples is greater than a certain threshold value, the distance is marked as g, and the gravity center indexes of the two samples are positioned in the front h, both the two samples can be used as new clustering gravity centers, namely new particle positions; wherein g takes the value of 3r, and h takes the value of 30%;

step 2.6F merges all the cluster centers found in step 2.6E, specifically: if the two nearest clustering centers are merged, the new clustering center after merging is the mean value of the two, and so on until M clustering centers are reached, namely the corrected particle position;

step 2.8, recalculating the fitness function value of the corrected particles according to the step 2.5;

step 2.9, calculating the global optimum value gbest of the particle swarm at the current time t, specifically: selecting the global optimal particle with the maximum fitness function value, wherein the corresponding position of the global optimal particle is the position of the global optimal particle;

step 2.10, randomly disturbing the positions of the particles by using Tent chaotic search to generate a chaotic sequence, and generating new particles according to the generated chaotic sequence, wherein the specific steps are as follows;

step 2.10A, normalizing the current global optimal value gbest from the optimized variable value interval [ Mmin, Mmax ] to the value interval [0,1] of the chaotic variable Zn;

step 2.10B, performing J-times chaotic disturbance on the global optimal value to generate a chaotic sequence, wherein the calculation mode is as follows:

step 2.10C, new particles are generated according to the generated chaos sequence, the particles are equal to the global optimum gbest plus k times of the chaos variable, the value of k is 0.05, namely X is X ═ X_gbest+k*z；

Step 2.11, calculating a fitness function value of the chaotic sequence according to the step 2.5, and selecting the chaotic sequence with the maximum fitness function value to replace the particle individuals which are optimal and not global optimal in the current particle swarm;

step 2.12, updating the acceleration coefficients c1 and c2 of the particle swarm, wherein the calculation mode is as follows:

where t is the number of iterations, i.e. time t, t_maxIs the maximum number of iterations;

step 2.14A, updating the particle swarm speed in the following manner:

V_i(k+1)＝w*V_i(k)+c₁*rand*(X_pbest,i-X_i)+c₂*rand*(X_gbest,i-X_i) (7)

wherein X_pbestAnd X_gbestRespectively representing the better value of the k generation individual and the best value in the solution group; c and w are aboveThe values of the acceleration coefficient and the inertial weight obtained in the steps 2.12 and 2.13; and 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 the particle position point pointing to the best position point of the group at the time of t-1, namely the global optimal value gbest, and reflects the cooperative cooperation and knowledge sharing among the particles;

step 2.14B, updating the positions of the particle swarms in the following manner:

X_i(k+1)＝X_i(k)+V_i(k+1) (8)

this expression indicates that the update of the current time t position of the particle is related to the position at the previous time, i.e., time t-1, and the velocity at the current time t.

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;

step 2.15, calculating the variance of the population fitness function value;

step 2.16, judging whether an iteration termination condition is reached, specifically: comparing whether the variance is smaller than a threshold value, wherein the threshold value is marked as v, and the value of v is 0.6 times of the fitness variance of the population during the first iteration; or whether t is greater than the maximum iteration time Tmax is compared.

If the two conditions are only one, the position of the particle is used as a clustering center for sorting the radar signals; if neither is satisfied, update t: t is t +1, and the step 2.4 is skipped;

and 3, evaluating the clustering result obtained in the step 2 based on the evaluation index, and regarding the data set D ═ x₁,x₂,...,x_mSuppose that clusters obtained by clustering are divided into Ω ═ ω { [ ω ]₁,ω₂,...,ω_KThe cluster given by the reference model is divided into C ═ C₁,c₂,...,c_KK is the number of clusters;

step 3.1 computing clustering quality CQ

Wherein N is_iIs omega_iIs under class c_iI.e. a homogeneity measure. When ω is_iWhen only samples from the same category are included, CQ is 1, so that the larger the clustering quality index is, the better the clustering effect is, and the higher the accuracy is.

Step 3.2, calculating and adjusting the landed index ARI;

let λ, λ^*Respectively representing cluster mark vectors corresponding to omega and C, and defining the paired samples

a＝|SS|,SS＝{(x_i,x_j)|λ_i＝λ_j,λ^* _i＝λ^* _j，i＜j},

b＝|SD|,SD＝{(x_i,x_j)|λ_i＝λ_j,λ^* _i≠λ^* _j，i＜j},

c＝|DS|,DS＝{(x_i,x_j)|λ_i≠λ_j,λ^* _i＝λ^* _j，i＜j},

d＝|DD|,DD＝{(x_i,x_j)|λ_i≠λ_j,λ^* _i≠λ^* _j，i＜j},

Wherein, ARI value range is [ -1,1 ]. In a broad sense, the ARI measures the degree of coincidence of two data distributions, and a larger value means that a clustering result is more coincident with a real situation.

Step 3.3 calculating normalized mutual information NMI

The normalized mutual information NMI calculation formula is as follows:

wherein I represents an increase amount of the category information Ω or a decrease amount of the uncertainty thereof on the premise of the given category cluster information C. P (w)_k),P(c_j),P(w_k∩c_j) Can be respectively regarded as that the samples belong to the cluster w_kBelong to the class c_jThe probability of belonging to both simultaneously. A larger value indicates a higher degree of similarity with the real category information.

Step 3.4 calculating centroid index CI

The centroid index is a measure of the difference between the two classes by clustering the centroids, and is calculated as follows:

CI＝max{CI(Ω,C),CI(C,Ω)} (12)

where CI ═ 0 indicates that the two clusters have the same structure, and a larger value indicates that the number of clusters assigned differently is larger, and the clustering effect is worse.

Step 3.5 calculating the Davison baudin index DBI

The index considers the average ratio of closeness to isolation in all clusters, and the calculation formula is as follows:

wherein e is_iAnd e_jIs the mean Euclidean distance of all samples i and j to the respective centroid, d_i,jIs the distance between the centroids. The smaller the DBI, the better the clustering effect.

Step 3.5 calculate Profile index SI

The index is defined as:

wherein a (i) is the average Euclidean distance of the sample i to the rest samples in the same cluster; b (i) is the minimum average euclidean distance between sample i and the samples in the remaining cluster. The higher the SI, the better the clustering scheme.

And 4, simulating other improved particle swarm optimization methods for comparison in order to illustrate the effectiveness of the method, wherein the improved particle swarm optimization methods are respectively a PSO method, a DPSOK method, an MfPSO method and an IPK-means method. For the several methodsAnd (3) similarly calculating various clustering evaluation indexes related to the step (3) by using the preprocessed data. In the experiments, the parameters involved included: maximum number of iterations t_max(ii) a The population scale P; maximum velocity v of particles_max(ii) a An inertial weight w; coefficient of acceleration c₁And c₂(ii) a In order to ensure the effect of each method, the specific parameter settings related to each method are shown in the following table 1 according to the parameter descriptions and simulation experiments mentioned in the literature:

TABLE 2 parameter settings

So far, from step 1 to step 4, the radar signal sorting method based on the dynamic modified chaotic particle swarm optimization is completed.

Fig. 3 is a graph of clustering results obtained using the DMCPSO method, showing the effect of radar signal characteristics on sorting results under different overlaps. The dots represent misclassified data.

In order to further show the superiority of the method, the iterative convergence time of different methods is analyzed, see table 3, and a curve diagram of the variance value of the fitness along with the variation of the iteration times is drawn, see fig. 4:

TABLE 3 run time of method

	PSO	DPSOK	MfPSO	IPK-Means	DMCPSO
						Run time/s	0.79	0.45	0.39	1.54	0.33

As can be seen from table 3, the DMCPSO method has the fastest convergence rate and the smallest stable variance value, which is obviously superior to other methods, and indicates that the population can always find the optimal position quickly in the iterative process, thereby converging to a satisfactory fitness value.

To further illustrate the effectiveness of the method, the method will be validated with several cluster sort indices. Specific results are shown in table 4:

as can be seen from Table 4, although the K-means method is simple and has fast convergence, the accuracy, matching degree and other aspects of the K-means method are different from those of the PSO series methods, which indicates that the PSO can well cope with complex radar data. Compared with other latest improved PSO methods, the DMCPSO method has obvious advantages on various indexes: the CQ value of the DMCPSO method reaches 94.99 percent, which shows that the clustering precision is highest, the classified data most tend to label data, and the influence of discrete data is hardly caused; the ARI value is maximum, which indicates that the clustering result is more consistent with the real situation and the probability of correct decision is maximum; the NMI value is maximum, which shows that the uncertainty of the category information is lower, and the relationship between the category data and the label data is closer; CI is reduced to 0, which shows that the classified data and the label data have better matching and have the same structure; the SI is maximum, which shows that the aggregation and separation between the clustered samples have higher values; the DBI is minimum, which shows that the intra-class distance is minimum and the inter-class distance is maximum after clustering.

The simulation results prove that the DMCPSO method has the best sorting effect under the condition of radar simulation data with higher pulse overlapping degree and less partial pulse number from various angles.

FIG. 2 is a characteristic space distribution diagram of 8 radar signals constructed by the radar signal sorting method based on dynamic modified chaotic particle swarm optimization; in the figure, the feature distribution of the radar pulse signal under a complex scene is shown from three dimensions of carrier frequency, arrival angle and pulse width, so that the signal features are obviously overlapped in space, and a good effect cannot be obtained by a traditional clustering method.

FIG. 3 is a clustering result diagram of the radar signal sorting method based on dynamic modified chaotic particle swarm optimization according to the invention; it can be seen from the figure that even 8 radar signal data with overlapped multi-parameter features under a complex background exist, the method can still accurately cluster and is excellent in effect.

FIG. 4 is a diagram of a comparison between a fitness variance convergence curve of the radar signal sorting method based on dynamic modified chaotic particle swarm optimization and other existing optimization clustering methods; it can be seen from the figure that the fitness variance of all the methods is decreasing with the increase of the number of iterations, and the change of the fitness value gradually decreases and becomes stable. The method finds the optimum by continually converging in the iterative process. However, the method provided by the application has the fastest convergence speed and the minimum stable variance value, is obviously superior to other methods, and shows that the group can always quickly find the optimal position in the iteration process, so that the group can converge to a satisfactory fitness value.

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. A chaotic particle swarm radar signal sorting method based on dynamic correction is characterized in that: the method comprises the following steps:

step 1 specifically comprises the following substeps:

step 1.1, constructing a radar data set;

wherein the dimension of the constructed radar data set is D;

the type of the radar signals in the radar data set in the step 1.1 is M;

step 2, sorting the radar signals and generating a clustering center for sorting the radar signals, namely a sorted result, and the method comprises the following substeps:

step 2.2, initializing the position of the M multiplied by D dimensional particles by using a Maxmin distance principle to obtain an initial clustering center of the population;

wherein, the iteration loop variable t represents t moment;

step 2.5, calculating a fitness function value of each particle;

wherein, the value range of a is 0 to 1;

step 2.6E, performing class separation inside each cluster, specifically: if the distance between the two samples is greater than a certain threshold value, the distance is marked as g, and the gravity center indexes of the two samples are positioned in the front h, the two samples are taken as a new clustering gravity center, namely a new particle position;

step 2.15, calculating the variance of the population fitness function value;

the evaluation indexes comprise clustering quality, adjusted lander indexes, normalized mutual information, centroid indexes, Davison Castle indexes and outline indexes.

2. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 1, wherein: in step 1, radar pulse PDW stream data is preprocessed, and the preprocessing comprises constructing a radar data set, collecting data to be sorted and revising the data to be sorted.

3. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 2, wherein: the radar data set in step 1.1 includes generating different kinds of radar signals to form data to be sorted and each kind of radar signal is a set of PDW stream data.

4. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 3, wherein: step 2, specifically: and (3) searching a global optimal solution, namely the sorted data clustering center, for the revised to-be-sorted data generated in the step 1.3 by adopting a chaotic particle swarm optimization method based on dynamic correction.

5. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 4, wherein: step 2.2, comprising the following substeps:

and 2.2C, selecting the point with the maximum closest distance from the first clustering center to the second clustering center as the central point of the third initial cluster, and repeating the steps until M initial cluster central points are selected and are used as the initial clustering centers of the population.

6. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 5, wherein: step 2.5, specifically:

and 2.5, dividing E1 by M, E and D to obtain a fitness function value at the time t.

7. The signal sorting method based on the dynamically modified chaotic particle swarm radar as claimed in claim 6, wherein: in step 2.16, the threshold v is set to range from 0.2 to 0.8 times the fitness variance of the population at time t ═ 1.