A distributed direct positioning method based on sensor networks

The distributed direct positioning method enhances accuracy and reduces costs by using a particle swarm and clustering algorithm with a precise first-order approach, addressing low signal-to-noise ratio challenges.

CN115119142BActive Publication Date: 2024-12-06HANGZHOU DIANZI UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202210713276.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-22
Publication Date
2024-12-06
Estimated Expiration
2042-06-22

AI Technical Summary

Technical Problem

Existing distributed direct positioning technologies suffer from low positioning accuracy and practical value in low signal-to-noise ratio environments, and traditional two-step methods lose target information due to intermediate parameter estimation.

Method used

A distributed direct positioning method using a particle swarm optimization algorithm, clustering algorithm, and precise first-order algorithm to estimate target position, eliminating the need for a fusion center and reducing computational and communication costs.

Benefits of technology

Improves positioning accuracy by estimating target position iteratively, achieving results comparable to centralized methods while reducing costs and maintaining robustness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115119142B_ABST
    Figure CN115119142B_ABST
Patent Text Reader

Abstract

The present invention discloses a distributed direct positioning method based on a sensor network. The current distributed direct positioning algorithm has the problem of low positioning accuracy under low signal-to-noise ratio. In order to improve the distributed direct positioning accuracy, the present invention adopts the following scheme: First, a high-resolution direct positioning distributed optimization model based on a subspace data fusion algorithm is derived. Second, in order to solve the problem of solving the initial value of the target position, a particle swarm algorithm based on vector evaluation is used to obtain a rough estimate of the iterative initial value of the target position, and then the traditional clustering algorithm is used to obtain the iterative initial value of the target position, thereby avoiding the problem of parameter matching of the traditional two-step positioning algorithm. Third, an accurate first-order algorithm in the field of distributed optimization is introduced, and the accuracy of centralized direct positioning is achieved through a two-step gradient information iterative solution, thereby solving the problem of positioning accuracy loss caused by distributed processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of passive positioning, and in particular relates to a distributed direct positioning method based on a sensor network. Background Art

[0002] Passive positioning technology has good positioning performance and security, and is widely used in many fields such as passive radar, acoustic sensors, sensor networks, etc. According to the positioning steps, passive positioning technology can be divided into traditional two-step method and direct positioning method. The traditional two-step method first estimates the positioning related parameters, such as direction angle, Doppler, delay parameters, etc. based on the sensor receiving signal, and then uses the positioning parameter information and sensor position information to jointly obtain the positioning related equations to estimate the target position. The two-step method requires the estimation of intermediate parameters, resulting in the loss of target information, so its positioning performance is suboptimal. The direct positioning method directly uses sensor signals to estimate the target position, which has higher positioning accuracy and robustness.

[0003] With the development of sensor network technology and the advent of the big data era, distributed positioning technology has become one of the current research hotspots. Combining direct positioning technology with distributed processing information technology, distributed direct positioning methods have been proposed one after another. Distributed direct positioning methods process sensor information in a distributed manner at each node, and have many advantages over centralized direct positioning methods. Distributed direct positioning methods do not have a fixed fusion center, and all sensors jointly undertake the positioning task, which is conducive to reducing the computational and communication costs of each sensor, while improving the robustness and scalability of the sensor network positioning system.

[0004] The existing distributed direct positioning technology has low positioning accuracy and low practical value in low signal-to-noise ratio environments. Therefore, this paper proposes a distributed direct positioning method based on sensor networks. The algorithm is based on the information of adjacent sensors. First, a particle swarm optimization algorithm based on vector evaluation is used to calculate a rough estimate of the target position. Then a clustering algorithm is used to obtain the initial value of the target position iteration. Finally, an accurate first-order algorithm is used to iteratively solve the target position, which greatly improves the positioning accuracy and has high practical application value. Summary of the invention

[0005] The purpose of the present invention is to provide a distributed direct positioning method based on a sensor network in view of the deficiencies of the prior art.

[0006] A distributed direct positioning method based on a sensor network comprises the following steps:

[0007] Step (1), generating a sensor receiving signal according to the simulated positioning scenario:

[0008] According to the preset radiation source target number Q, snapshot number N, carrier frequency fc, modulation frequency u, sampling rate fs, the discrete form s of the linear frequency modulation signal is generated. q (n), n=1,...,N, q=1,...,Q;

[0009] Preset the number of sensor array elements M and the array element spacing d, and get the mth array element vector Where, m = 1, ..., M, T represents transposition;

[0010] According to the sensor position p r , radiation source target position p q , signal wavelength λ, and obtain the wave number vector information Among them, ||·|| is the modulus symbol, Δdist represents the distance vector from the sensor position to the radiation source target position, and u represents the unit direction vector from the radiation source target to the sensor;

[0011] The array element vector and wave number vector are combined to form the sensor r array steering vector:

[0012]

[0013] Where j represents an imaginary number;

[0014] Call the randn() function and set the appropriate signal-to-noise ratio parameter SNR to add noise w r (n), then the sensor r array receives the signal in the following form:

[0015]

[0016] Among them, s q (n) represents the discrete form of the radiation source target q signal;

[0017] For ease of calculation, assume that the N snapshot arrays of sensor r receive signal matrix Z r =[z r (1),...,z r (N)], the array steering vector matrix A of the Q radiation source targets of sensor r r =[a r (p1),...,a r (p Q )], Q radiation source target discrete signal matrix S r =[s1,...,s q ] T , where s q =[s q (1),...,s q (N)] T , sensor r array receives noise signal matrix, Wr =[w r (1),..,w r (n)]; convert the sensor r array receiving signal model into a matrix form:

[0018] Z r =A r S r +W r (3)

[0019] Step (2), each sensor calculates the covariance matrix of the array received signal and constructs the subspace cost function J r (p); specifically:

[0020] According to the array received signal of sensor r, the covariance matrix of the array received signal is obtained:

[0021]

[0022] Where Z r represents the array receiving signal of sensor r, and the symbol H represents the conjugate transpose operation;

[0023] Perform eigenvalue decomposition according to formula (4) to obtain M eigenvalues ​​eig and their corresponding M eigenvectors ξ;

[0024] The above eigenvalues ​​are sorted by size to satisfy the following relationship:

[0025] eig1≥eig2≥…eig Q >eig Q+1 =…=eig M (5)

[0026] Take the eigenvectors corresponding to MQ smaller eigenvalues ​​to form the noise subspace According to the principle of MUSIC spatial spectrum estimation algorithm, the subspace cost function of sensor r is constructed as:

[0027]

[0028] Where p is the possible location of the target, a r (p) represents the array steering vector corresponding to position p, when p = p q When , the subspace cost function reaches a minimum value;

[0029] Step (3): Combine the sensor network structure to construct a distributed direct positioning local cost function J′ r (p); specifically:

[0030] Each sensor exchanges subspace cost function information with adjacent sensors to construct a distributed direct positioning local cost function:

[0031]

[0032] in Among them C r represents the sensor set consisting of sensor r and its adjacent sensors, denoted as sensor cluster, D r Represents sensor cluster C r Number of internal sensors, J l (p) represents the subspace cost information of sensor l, l is the sensor cluster C corresponding to sensor r r Sensors that may be obtained in

[0033] Step (4), using a particle swarm optimization algorithm based on vector evaluation to estimate the rough estimate of the initial value of the target position iteration;

[0034] Step (5), estimating the initial value of target position iteration based on cluster optimization algorithm;

[0035] Step (6), optimize the target positioning result based on the precise first-order algorithm.

[0036] Preferably, step (4) is to use each sensor as an independent intelligent agent, each intelligent agent uses the local direct positioning cost function as the fitness function, create a particle swarm, and parallelly calculate and estimate the target position, which is specifically as follows:

[0037] 4-1 Define and initialize the number of particles N of each sensor particle swarm algorithm p , the number of iterations N iter1 , optimal particle position in the sensor cluster The optimal particle position p of the particle swarm in each sensor g[r] , the historical optimal position p of particle i in the particle swarm i , the current position xoy of particle i in the particle swarm i ;

[0038] 4-2 Update the historical optimal position of each particle i:

[0039] If J′ r (xoy i ) <J′ r (p i ), let p i =xoy i , otherwise it is not updated; among them, J′ r (xoy i ) represents the distributed direct positioning local cost function J′ r (p), p takes xoy i The value at time, J′ r (p i ) represents the distributed direct positioning local cost function J′r (p), p takes p i The value at time;

[0040] 4-3 Update the optimal particle position of each particle group:

[0041] If J′ r (p i ) <J′ r (p g[r] ), let p g[r] =p i , otherwise it is not updated; among them, J′ r (p g[r] ) represents the distributed direct positioning cost function J′ r (p), p takes p g[r] The value at time;

[0042] 4-4 Update the best particle position in the sensor cluster:

[0043] Randomly select the optimal particle position of the particle swarm of sensor l in the sensor cluster, that is, any optimal particle position p of the sensor g[l] ,l∈C r Assign the best particle position to the sensor cluster where the sensor belongs

[0044] 4-5 Update the velocity status of each particle:

[0045]

[0046] Among them, v i is the velocity of the ith particle, xoy i is the current position of the ith particle, χ represents the scaling factor, and Respectively expressed in and Random vectors generated in the range Represents the vector dot product operator;

[0047] 4-6 Update the current position status of each particle:

[0048] xoy i =xoy i +v i (9)

[0049] 4-7 Repeat the above steps (4-2 to 4-6) until the iteration reaches the preset value, and let the historical optimal position p of each particle i Serves as a rough estimate of the initial value for the target position iteration.

[0050] Preferably, step (5) is to use clustering to accelerate the iterative initial value of the target position, as follows:

[0051] 5-1 Based on the target number, the number of clusters of the preset clustering algorithm is loc_num;

[0052] 5-2 Roughly estimate p based on the iterative initial value obtained in step (4) i , that is, the historical optimal position information of each particle, randomly determine the historical optimal position of loc_num particles as the centroid of the clustering algorithm;

[0053] 5-3 For each sensor, there is step (4) where each particle has the historical optimal position p i Calculate the distances to loc_num centroids in turn, select the centroid with the smallest distance, and assign it to the cluster corresponding to the centroid, until all particles in the sensor are assigned to loc_num clusters, and update the centroid of loc_num clusters to the average value of all points in the cluster;

[0054] 5-4 Repeat the above step (5-3) until all particle allocation results no longer change, take the centroid position of loc_num clusters as the initial value of the target position iteration, and solve the precise value of the estimated target position for loc_num generations of parallel iterations through step (6).

[0055] Preferably, step (6) is based on the multi-target iterative initial value obtained in steps (4)-(5), and an accurate first-order algorithm is used in parallel to improve the distributed direct positioning accuracy, specifically:

[0056] 6-1 In order to obtain the gradient information of the distributed direct positioning local cost function First, the complex information contained in J′(p) is converted into real-valued information:

[0057] y r (p) = [Re(a r (p)) T Im(a r (p)) T ] T

[0058]

[0059] Where Re(a r (p)) represents the real part of the sensor r array guidance vector information when the target position is p, Im(a r (p)) represents the imaginary part of the sensor r array guidance vector when the target position is p, y r (p) represents the real-valued information of the sensor r array guidance vector when the target position is p, represents the real part of the noise subspace information of sensor r, represents the imaginary part of the noise subspace of sensor r, is the intermediate variable of mathematical operation, represents the real-valued information of the noise subspace covariance matrix of sensor r;

[0060] 6-2 According to the above definition, the gradient of the local cost function of distributed direct positioning is calculated as:

[0061]

[0062] in, represents the real-valued information of the noise subspace covariance matrix of sensor l, where l is the sensor cluster C corresponding to sensor r r The sensors that may be obtained in Represents the gradient information of the sensor r array steering vector;

[0063] The gradient information of the sensor r array steering vector is specifically:

[0064]

[0065] Among them, the gradient information of the wave number vector of sensor r at position p is:

[0066]

[0067] Where I2 represents the second-order identity matrix;

[0068] 6-3 Each sensor performs initialization iteration, and the estimated target position of each sensor in the first iteration is:

[0069]

[0070] where a r,l represents the weighting coefficient, α represents the fixed step size of the gradient algorithm, and p l,1 represents the initialization iteration value of sensor l in the sensor cluster, p r,1 Represents the initialization iteration value of sensor r in the sensor cluster;

[0071] 6-4 Each sensor estimates the target position in parallel iterations, and the target position estimate of each sensor is:

[0072]

[0073] Where k represents the number of iterations, p l,k represents the k-1th iteration value of sensor l in the sensor cluster, p r,k represents the k-1th iteration value of sensor r in the sensor cluster;

[0074] 6-5 Based on the first iteration target position estimation value obtained in step (6-3), each sensor repeats step (6-4) using two-step gradient information until the number of iterations reaches a preset value or the iteration error of the target position estimation value is less than a set value, then the target position estimation value in the current iteration is the final positioning result of the sensor;

[0075] Preferably, if there are multiple targets, steps (6-3 to 6-4) are repeated in parallel based on different target iteration initial values.

[0076] A second object of the present invention is to provide an electronic device, comprising a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the described method.

[0077] A third object of the present invention is to provide a machine-readable storage medium storing machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the described method.

[0078] The present invention has the following beneficial effects:

[0079] 1. Based on the subspace direct positioning algorithm, the present invention derives the local cost function of distributed direct positioning, adopts a distributed optimization method to estimate the target position, does not require a fusion center, and reduces the communication cost of each sensor transmitting data.

[0080] 2. The present invention adopts a particle swarm algorithm based on vector evaluation and a clustering algorithm to estimate the initial value of the target position iteration, thus avoiding the problem of target parameter association in the traditional two-step positioning algorithm.

[0081] 3. The present invention adopts a distributed precise first-order algorithm to estimate the target position, and approaches the accuracy of centralized direct positioning through a two-step gradient information iteration solution, which is helpful to solve the problem of positioning accuracy loss caused by distributed processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is the overall flow chart of the present invention;

[0083] Figure 2 A sensor network topology diagram of the present invention;

[0084] Figure 3 It is a flow chart of the particle optimization algorithm based on vector evaluation of the present invention;

[0085] Figure 4 It is a flow chart of the clustering algorithm of the present invention;

[0086] Figure 5 It is a framework diagram of the precise first-order algorithm in the present invention;

[0087] Figure 6 This is a simulation result diagram of rough estimation of initial value of target position iteration based on particle swarm algorithm in the present invention;

[0088] Figure 7 This is a simulation result diagram of the target position iteration initial value estimation based on the clustering algorithm in the present invention;

[0089] Figure 8 This is a diagram of the simulation results of the distributed direct positioning performance of the present invention. DETAILED DESCRIPTION

[0090] The present invention will be further described below in conjunction with the accompanying drawings.

[0091] like Figure 1 As shown in Figure 1, a distributed direct positioning technology method based on sensor networks is shown in Figure 1. Figure 1 The specific steps are as follows:

[0092] Step (1), the present invention realizes target positioning based on a sensor network, and its network topology is as follows Figure 2 As shown, first generate the sensor simulation receiving signal, the method is as follows:

[0093] The preset radiation source target number Q = 3, the snapshot number N = 200, the carrier frequencies of the three target signals are fc = 15MHz, 75MHZ, 150MHz, the sampling rate fs = 500MHz, the modulation rate u, and the discrete form s of the linear frequency modulation signal is generated. q (n), n=1,...,N, q=1,...,Q;

[0094] According to the sensor position p r , radiation source target position p q , signal wavelength λ, and obtain the wave number vector information

[0095] Assume the number of sensor array elements M = 8 and the array element spacing Get the array element vector

[0096] The array element vector and wave number vector are combined to form the traditional array steering vector matrix as shown in formula (1):

[0097]

[0098] Call the randn() function and set the appropriate signal-to-noise ratio parameter SNR to add noise w r (n), sensor r receives data in the following form:

[0099]

[0100] For ease of calculation, let the N snapshot received signal matrix Z of sensor r be r =[z r (1),...,z r (N)], the Q array steering vector matrix A of sensor r r =[a r (p1),...,a r (p Q )], Q radiation source target discrete signal matrix S r =[s1,...,s q ] T , where s q =[s q (1),...,s q (N)] T , sensor r array receives noise signal matrix, W r =[w r (1),..,w r (n)]. The sensor r array receiving signal model is converted into a matrix form, as shown in formula (3):

[0101] Z r =A r S r +W r (3)

[0102] Step (2), each sensor calculates the covariance matrix of the received signal and constructs the subspace cost function J r (p);

[0103] According to the array receiving signal of sensor r, the covariance matrix of its array receiving signal can be obtained:

[0104]

[0105] Where Z r represents the received signal of array r, and the symbol H represents the conjugate transpose operation;

[0106] By performing eigenvalue decomposition according to formula (4), we can obtain M eigenvalues ​​eig and their corresponding M eigenvectors ξ. The eigenvalues ​​are sorted in order to satisfy the following relationship:

[0107] eig1≥eig2≥…eig Q >eig Q+1 =…=eig M (5)

[0108] Take the eigenvectors corresponding to MQ smaller eigenvalues ​​to get the noise subspace According to the principle of MUSIC spatial spectrum estimation algorithm, the subspace cost function of sensor r is constructed as:

[0109]

[0110] Step (3), combined with the sensor network topology, such as Figure 2 As shown, construct a distributed direct positioning local cost function J′ r (p);

[0111] Each sensor exchanges subspace cost function information with adjacent sensors to construct a distributed direct positioning local cost function:

[0112]

[0113] In the formula Among them C r represents the sensor set consisting of sensor r and its adjacent sensors, D r Indicates the number of sensors in the sensor cluster.

[0114] Step (4), such as Figure 3 The particle swarm optimization algorithm based on vector evaluation is used to estimate the rough estimate of the initial value of the target position iteration; the sub-steps of this method are as follows:

[0115] 4-1 Take each sensor as an independent agent, and each agent uses the local direct positioning cost function as the fitness function to create a particle swarm. Each sensor initializes the particle swarm algorithm parameters and the number of particles N p =20, number of iterations N iter1 = 50. Defines the optimal particle position in the sensor cluster The optimal particle position p of the particle swarm in each sensor g[r] , the historical optimal position p of particle i in the particle swarm i , the current position xoy of particle i in the particle swarm i .

[0116] 4-2 Update the historical optimal position of each particle. If J′ r (xoy i ) <J′ r (p i ), let p i =xoy i ;

[0117] 4-3 Update the optimal position of each particle group. If J′ r (p i ) <J′ r (p g[r] ), let p g[r] =p i ;

[0118] 4-4 Update the best particle position in the sensor cluster: Randomly take the best particle position of the particle group of sensor l in the sensor cluster, that is, any p g[l] ,l∈C r , assigned to

[0119] 4-5 Update the velocity status of each particle:

[0120]

[0121] 4-6 Update the current position status of each particle:

[0122] xoy i =xoy i +v i (9)

[0123] Among them, v i is the current speed of the ith particle, xoy i is the current position of the ith particle. χ represents the scaling factor, and Respectively expressed in and Random vectors generated in the range represents the vector dot product operator. Set χ to 0.7298,

[0124] 4-7 Repeat the above steps (4-2 to 4-6) until the iteration reaches the preset value, and let the historical optimal position p of each particle i As a rough estimate of the initial value for the iteration.

[0125] Step (5), such as Figure 4 Adjacent sensors share particle information in their particle swarms, and then use a clustering optimization algorithm to estimate the initial value of the target position iteration; the specific sub-steps are as follows:

[0126] 5-1 Based on the target number, the number of clusters of the preset clustering algorithm is loc_num;

[0127] 5-2 Roughly estimate p based on the iterative initial value obtained in step (4) i , that is, the historical optimal position information of each particle, randomly determine the historical optimal position of loc_num particles as the centroid of the clustering algorithm;

[0128] 5-3 For each sensor, there is step (4) where each particle has the historical optimal position p iCalculate the distances to loc_num centroids in turn, select the centroid with the smallest distance, and assign it to the cluster corresponding to the centroid, until all particles in the sensor are assigned to loc_num clusters, and update the centroid of loc_num clusters to the average value of all points in the cluster;

[0129] 5-4 Repeat the above step (5-3) until all particle allocation results no longer change, take the centroid position of loc_num clusters as the initial value of the target position iteration, and solve the precise value of the estimated target position for loc_num generations of parallel iterations through step (6).

[0130] Step (6), such as Figure 5 Based on the initial value of the multi-target position iteration obtained in step 5, the precise first-order algorithm is used to optimize the target positioning result and solve the multi-target positioning result by parallel iteration. Specifically:

[0131] 6-1 In order to obtain the gradient information of the distributed direct positioning local cost function This step first converts the complex information contained in J′(p) into real-valued information:

[0132] y r (p) = [Re(a r (p)) T Im(a r (p)) T ] T

[0133]

[0134] Among them, Re(a r (p)) represents the real part of the sensor r array guidance vector information when the target position is p, Im(a r (p)) represents the imaginary part of the sensor r array steering vector when the target position is p, and y(p) represents the real value information of the array steering vector when the target position is p. represents the real part of the noise subspace information of sensor r, represents the imaginary part of the noise subspace of sensor r, It is an intermediate variable in mathematical operations and has no specific physical meaning. represents the real-valued information of the noise subspace covariance matrix of sensor r;

[0135] 6-2 According to the above definition, the gradient of the local cost function of distributed direct positioning is calculated as:

[0136]

[0137] in, represents the real-valued information of the noise subspace covariance matrix of sensor l, where l is the sensor cluster C corresponding to sensor r r The possible sensors in the r , Represents the gradient information of the sensor r array steering vector;

[0138] The gradient information of the sensor r array steering vector is specifically:

[0139]

[0140] Among them, the gradient information of the sensor r wave number vector is:

[0141]

[0142] Where I2 represents the second-order identity matrix;

[0143] 6-3 Each sensor performs initialization iteration, and the estimated target position of each sensor in the first iteration is:

[0144]

[0145] The fixed step size α of the gradient algorithm is 5, and the weight coefficient a is r,l for:

[0146]

[0147] where a r,k Represents sensor cluster C r Weighting coefficient of internal sensor m, C r represents the sensor set consisting of sensor r and its adjacent sensors, D r Represents sensor cluster C r Number of internal sensors;

[0148] 6-4 Each sensor estimates the target position in parallel iterations, and the target position estimate of each sensor is:

[0149]

[0150] Where k represents the number of iterations, p l,k represents the k-1th iteration value of sensor l in the sensor cluster, p r,k represents the k-1th iteration value of sensor r in the sensor cluster;

[0151] 6-5 Based on the first iterative target position estimate obtained in step (6-3), each sensor repeats step (6-4) using two-step gradient information until the number of iterations reaches a preset value or the iterative error of the target position estimate is less than the set value. The target position estimate in the current iteration is the final positioning result of the sensor. If there are multiple targets, steps (6-3 to 6-4) are repeated in parallel based on different target iteration initial values.

[0152] The present invention is based on the distributed positioning framework of sensor networks. Firstly, a particle optimization algorithm based on vector evaluation is used to estimate a rough estimate of the initial value of the target position iteration. Then, a clustering algorithm is used to estimate the initial value of the target position iteration. Finally, an accurate first-order algorithm is used based on the target iteration initial value to optimize the target positioning estimation result, thereby improving the accuracy of distributed positioning, and the effectiveness of the algorithm is proved by simulation experiments.

[0153] Based on the above simulation parameters, the present invention is simulated and analyzed, and the specific results are as follows:

[0154] Figure 6 The simulation result diagram of the rough estimation of the initial value of the target position iteration of the particle swarm algorithm based on vector evaluation in the present invention, the simulation parameters are set as follows: there are three target radiation source signals, the signal-to-noise ratio is set to 10dB, and the radiation source target positions are set to x1=5000, y1=5000, x2=13000, y2=5000, x3=12000, y3=12200 (unit is m, the same below), as can be seen from the figure, the particle swarm optimization algorithm based on vector evaluation with 50 iterations has a more obvious clustering trend. Although the positioning accuracy of the particle swarm algorithm based on vector evaluation of the present invention is poor, it does not require parameter matching and can be used as a rough estimate of distributed direct positioning.

[0155] Figure 7 This is a simulation result diagram of the initial value estimation of the target position iteration based on the clustering algorithm in the present invention. As can be seen from the figure, using the clustering algorithm to correct the results of 50 iterations of the particle swarm algorithm based on vector evaluation can obtain a more reasonable initial positioning estimate in advance, thereby avoiding the problem of additional calculation caused by increasing the number of iterations of the particle swarm algorithm.

[0156] Figure 8This is a simulation result diagram of the distributed direct positioning performance of the present invention. Under the above-mentioned simulation parameters, 300 Monte Carlo simulation experiments were performed for each signal-to-noise ratio (SNR), and the mean square error (RMSE) was used as the positioning performance indicator. In order to compare the positioning performance of the method of the present invention, the present invention compares the positioning performance of the centralized positioning algorithm and the distributed gradient algorithm. It can be seen from the figure that the traditional distributed gradient algorithm has a loss in positioning accuracy compared to the centralized positioning algorithm under low signal-to-noise ratio. The positioning accuracy of the precise first-order algorithm used in this article can be close to that of the centralized positioning algorithm, and has a higher positioning accuracy.

Claims

1. A distributed direct positioning method based on sensor networks, characterized in that The following steps are involved: Step (1), generating a radiation source multi-array signal according to the simulation positioning scene: According to the preset radiation source target number Q, snapshot number N, carrier frequency fc, modulation frequency u, sampling rate fs, the discrete form s of the linear frequency modulation signal is generated. q (n), n=1,...,N, q=1,...,Q; Preset the number of sensor array elements M and the array element spacing d, and get the mth array element vector Where, m = 1, ..., M, T represents transposition; According to the sensor position p r , radiation source target position p q , signal wavelength λ, and obtain the wave number vector information Among them, ||·|| is the modulus symbol, Δdist represents the distance vector from the sensor position to the radiation source target position, and u represents the unit direction vector from the radiation source target to the sensor; The array element vector and wave number vector are combined to form the sensor r array steering vector: Where j represents an imaginary number; Call the randn() function and set the appropriate signal-to-noise ratio parameter SNR to add noise w r (n), then the sensor r array receives the signal in the following form: Among them, s q (n) represents the discrete form of the radiation source target q signal; For ease of calculation, assume that the N snapshot arrays of sensor r receive signal matrix Z r =[z r (1),...,z r (N)], the array steering vector matrix A of the Q radiation source targets of sensor r r =[a r (p1),...,a r (p Q )], Q radiation source target discrete signal matrix S r =[s1,...,s Q ] T , where s q =[s q (1),...,s q (N)] T , sensor r array receives noise signal matrix, W r =[w r (1),..,w r (n)]; convert the sensor r array receiving signal model into a matrix form: Z r =A r S r +W r (3) Step (2), each sensor calculates the covariance matrix of the array received signal and constructs the subspace cost function J r (p); specifically: According to the array received signal of sensor r, the covariance matrix of the array received signal is obtained: Where Z r represents the array receiving signal of sensor r, and the symbol H represents the conjugate transpose operation; Perform eigenvalue decomposition according to formula (4) to obtain M eigenvalues ​​eig and their corresponding M eigenvectors ξ; The above eigenvalues ​​are sorted by size to satisfy the following relationship: own1≥own2≥…own Q >own Q+1 =…=own M (5) Take the eigenvectors corresponding to MQ smaller eigenvalues ​​to form the noise subspace According to the principle of MUSIC spatial spectrum estimation algorithm, the subspace cost function of sensor r is constructed as: Where p is the possible location of the target, a r (p) represents the array steering vector corresponding to position p, when p = p q When , the subspace cost function reaches a minimum value; Step (3): Combine the sensor network structure to construct a distributed direct positioning local cost function J′ r (p); specifically: Each sensor exchanges subspace cost function information with adjacent sensors to construct a distributed direct positioning local cost function: in Among them C r represents the sensor set consisting of sensor r and its adjacent sensors, denoted as sensor cluster, D r Represents sensor cluster C r Number of internal sensors, J l (p) represents the subspace cost information of sensor l, l is the sensor cluster C corresponding to sensor r r Sensors that may be obtained in Step (4) uses a particle swarm optimization algorithm based on vector evaluation to estimate the rough estimate of the initial value of the target position iteration, which is as follows: 4-1 Define and initialize the number of particles N of each sensor particle swarm algorithm p , the number of iterations N iter1 , optimal particle position in the sensor cluster The optimal particle position p of the particle swarm in each sensor g[r] , the historical optimal position p of particle i in the particle swarm i , the current position xoy of particle i in the particle swarm i ; 4-2 Update the historical optimal position of each particle i: If J′ r (xoy i ) <J′ r (p i ), let p i =xoy i , otherwise it is not updated; among them, J′ r (xoy i ) represents the distributed direct positioning local cost function J′ r (p), p is xoy i The value at time, J′ r (p i ) represents the distributed direct positioning local cost function J′ r (p), p takes p i The value at time; 4-3 Update the optimal particle position of each particle group: If J′ r (p i ) <J′ r (p g[r] ), let p g[r] =p i , otherwise it is not updated; among them, J′ r (p g[r] ) represents the distributed direct positioning cost function J′ r (p), p takes p g[r] The value at time; 4-4 Update the best particle position in the sensor cluster: Randomly select the optimal particle position of the particle swarm of sensor l in the sensor cluster, that is, any optimal particle position p of the sensor g[l] ,l∈C r Assign the best particle position to the sensor cluster where the sensor belongs 4-5 Update the velocity status of each particle: Among them, v i is the velocity of the ith particle, xoy i is the current position of the ith particle, χ represents the scaling factor, and Respectively expressed in and Random vectors generated in the range Represents the vector dot product operator; 4-6 Update the current position status of each particle: hey i =oh i +v i (9) 4-7 Repeat the above steps 4-2 to 4-6 until the iteration reaches the preset value, and let the historical optimal position p of each particle i As a rough estimate of the initial value of the target position iteration; Step (5), estimating the initial value of the target position iteration based on the clustering optimization algorithm; the details are as follows: 5-1 Based on the target number, the number of clusters of the preset clustering algorithm is loc_num; 5-2 Roughly estimate p based on the iterative initial value obtained in step (4) i , that is, the historical optimal position information of each particle, randomly determine the historical optimal position of loc_num particles as the centroid of the clustering algorithm; 5-3 For each sensor, there is step (4) where each particle has the historical optimal position p i Calculate the distances to loc_num centroids in turn, select the centroid with the smallest distance, and assign it to the cluster corresponding to the centroid, until all particles in the sensor are assigned to loc_num clusters, and update the centroid of loc_num clusters to the average value of all points in the cluster; 5-4 Repeat the above step 5-3 until all particle allocation results no longer change, take the centroid position of loc_num clusters as the initial value of the target position iteration, and solve the precise value of the estimated target position for loc_num generations through parallel iteration of step (6); Step (6), optimizing the target positioning result based on the precise first-order algorithm, specifically: 6-1 In order to obtain the gradient information of the distributed direct positioning local cost function First, the complex information contained in J′(p) is converted into real-valued information: y r (p)=[Re(a r (p)) T Im(a r (p)) T ] T Where Re(a r (p)) represents the real part of the sensor r array guidance vector information when the target position is p, Im(a r (p)) represents the imaginary part of the sensor r array guidance vector when the target position is p, y r (p) represents the real-valued information of the sensor r array guidance vector when the target position is p, represents the real part of the noise subspace information of sensor r, represents the imaginary part of the noise subspace of sensor r, is the intermediate variable of mathematical operation, represents the real-valued information of the noise subspace covariance matrix of sensor r; 6-2 According to the above definition, the gradient of the local cost function of distributed direct positioning is calculated as: in, represents the real-valued information of the noise subspace covariance matrix of sensor l, where l is the sensor cluster C corresponding to sensor r r The sensors that may be obtained in Represents the gradient information of the sensor r array steering vector; The gradient information of the sensor r array steering vector is specifically: Among them, the gradient information of the wave number vector of sensor r at position p is: Where I2 represents the second-order identity matrix; 6-3 Each sensor performs initialization iteration, and the estimated target position of each sensor in the first iteration is: where a r,l represents the weighting coefficient, α represents the fixed step size of the gradient algorithm, and p l,1 represents the initialization iteration value of sensor l in the sensor cluster, p r,1 Represents the initialization iteration value of sensor r in the sensor cluster; 6-4 Each sensor estimates the target position in parallel iterations, and the target position estimate of each sensor is: Where k represents the number of iterations, p l,k represents the k-1th iteration value of sensor l in the sensor cluster, p r,k represents the k-1th iteration value of sensor r in the sensor cluster; 6-5 Based on the first iterative target position estimate obtained in step 6-3, each sensor repeats step 6-4 using two-step gradient information until the number of iterations reaches a preset value or the iterative error of the target position estimate is less than the set value. The target position estimate in the current iteration is the final positioning result of the sensor.

2. A distributed direct positioning method based on a sensor network according to claim 1, characterized in that If there are multiple targets, steps 6-3 to 6-4 are repeated in parallel based on different target iteration initial values.

3. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the method according to any one of claims 1 to 2.

4. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions prompt the processor to implement the method described in any one of claims 1-2.

Citation Information

Cited By

  • Multi-satellite direct positioning method based on CUDA (Compute Unified Device Architecture) parallel computing

    CN120103388A