RFID antenna optimization deployment method based on single-base communication and phased array

Abstract

An RFID antenna optimization deployment method based on single-base communication and a phased array specifically comprises the following steps: establishing a channel link model based on a linear phased array antenna, and accurately estimating receiving power between a reader and a tag; a half-wave dipole antenna is used as a tag antenna, and a gain expression of a one-dimensional linear phased-array antenna is derived; integrating the reader antenna and the tag gain model into the same Cartesian coordinate system; establishing a channel link model on the basis, and constructing an RNP problem optimization model in a single-base communication mode; and an improved sparrow search algorithm EGOSSA is put forward to solve the RNP problem, a final optimal deployment scheme is obtained through iterative operation, and the RFID antenna optimization deployment method based on single-base communication and the phased array is completed. According to the method, the coverage rate of the system to the tags is improved, communication interference between the readers and the tags is reduced, and the method has certain guiding significance in improving the quality and efficiency of solving the RNP problem.

Description

Technical Field

[0001] The invention belongs to the field of Internet of Things and relates to an RFID antenna optimization deployment method based on single-base communication and phased array. Background Art

[0002] With the advent of the Internet of Things, the interconnection of all things has become possible. Radio Frequency Identification (RFID) technology, as one of the important components of the Internet of Things technology, is widely used in intelligent transportation systems, access control, medical health, retail, logistics and supply chain management, etc., realizing real-time monitoring, automated production and inventory management, and bringing efficient, accurate and automated data management solutions to various industries. A typical RFID system usually consists of three parts: readers, tags and back-end systems. Actual large-scale RFID application scenarios usually require the deployment of a sufficient number of readers and antennas to achieve the identification and tracking of a large number of tags. Therefore, how to carry out effective network deployment has become a problem that needs to be solved urgently. This type of problem is called RFID Network Planning (RNP) problem.

[0003] When solving the RNP problem, the following two problems usually need to be considered: (1) How to deal with the load imbalance between different readers. Load imbalance will reduce the information transmission efficiency of the RFID system and consume a lot of time, thus affecting the coverage of the entire system; (2) How to balance the applicability and economy of the system. Although deploying too many readers meets the needs of tag identification, it also increases the economic cost. In general, the ultimate goal of solving the RNP problem is to optimize multiple objectives (tag coverage, interference between systems, deployment costs, and reader load balance, etc.) by adjusting the control variables of the reader (quantity, coordinates, and antenna parameters, etc.) to achieve the best deployment effect.

[0004] Therefore, in order to solve the above problems, the present invention proposes an RFID antenna optimization deployment method based on single-base communication and phased array. Summary of the invention

[0005] The purpose of the present invention is to provide an RFID antenna optimization deployment method based on single-base communication and phased array, so as to improve the system coverage of tags and reduce the communication interference between readers and tags. The modeling scheme adopted by the present invention is, firstly, to derive the gain expression of the one-dimensional linear phased array antenna based on the gain expression of the two-dimensional phased array antenna, and to establish a channel link model based on this, and then to establish an RNP problem optimization model under the single-base communication mode in combination with the optimization objectives such as tag coverage, positioning accuracy and system interference, and finally to propose an improved sparrow search algorithm (Enhanced SparrowSearch Algorithm based on Generative Adversarial Network and OrthogonalExperimental Design, EGOSSA) to solve the RNP problem. Specifically, the following steps are included:

[0006] Step 1: Establish a channel link model based on a linear phased array antenna to accurately estimate the received signal power in the forward link and reverse link between the reader and the tag. The Radio Frequency Identification (RFID) system usually adopts an RFID single base station system. The received signal power P of the tag in the forward link is r,T =10log 10 (ρ L P Tx G T G T L(d))+30, the received signal power of the reader antenna in the backward link where ρ L represents the polarization loss factor, R Tx Indicates the transmission power of the reader, G T and G R They represent the gains of the tag antenna and the reader antenna, μ T represents the power transmission efficiency, κ represents the modulation efficiency, Γ is the differential reflection coefficient of the tag, and L(d) represents the path loss factor, which is specifically expressed as L(d) = λ 2 / (16π 2 η 2 d 2 ), where d represents the distance between the reader and the tag, and η represents the path loss exponent;

[0007] Step 2: Use a half-wave dipole antenna as the tag antenna, and its gain can be expressed as G T (θ T ,φ T )=1.641(cos 2 (0.5πcos(θ T ))) / sin2 (θ T ), where the long side of the tag antenna is the z-axis and the short side is the x-axis, is the radiation direction, θ T and φ T represents a pair of directivity parameters, θ T express arrive The inclination angle, φ T express arrive The rotation angle projected on the XOY plane uses a one-dimensional linear phased array antenna as the reader antenna, and its gain can be expressed as G R (θ R ,φ R )=eD(θ R ,φ R ),in e represents the efficiency factor, D(θ R ,φ R ) represents the directivity coefficient, S(θ R ,φ R ) represents the planar array factor, I n They represent the excitation amplitude along the y-axis, n represents the array element index, and y n represents the coordinate of the element along the y-axis, α represents the phase difference of the excitation, and θ R and φ R Definition and θ T and φ T Similarly, let cosθ y = sinθ R cosφ R ,S(θ R ,φ R ) can be expressed as Among them, θ y Indicates the y-axis and The angle between the two, N represents the number of linear array elements, I n represents the excitation amplitude along the y-axis, y n represents the coordinate of the array element along the y-axis, α is the phase difference of the excitation, and a 1×4 linear phased array antenna is used. Assuming e=1, I n =1, k = 2π / λ, d = 0.5λ, the array element coordinates are (0, -0.75λ, 0), (0, -0.25λ, 0), (0, 0.25λ, 0), (0, 0.75λ, 0), the planar array factor S(θ y ) can be expressed as After formula derivation and definite integral calculation, the gain of the reader antenna can be expressed as

[0008] Step 3: Integrate the gain models of the reader antenna and the tag into the same Cartesian coordinate system. (x T ,y T , z T ) and (x R ,y R , z R ) represent the spatial coordinates of the tag antenna and the reader antenna, and They represent the positions of the reader antenna and the tag antenna respectively, where Indicates from arrive The inclination angle, Indicates from arrive The rotation angle mapped in the XOY plane, Representative from arrive The inclination angle, express arrive In the XOY plane mapping rotation angle, the reader antenna is usually located higher than the tag antenna and radiates at a downward angle, G T and G R Can be used and Represented as G T =1.641cos 2 (A T / B T ) / (1-(C T / D T ) 2 ), in

[0009] Step 4: Establish the RNP (RFID Network Planning) problem model, take the minimization of coverage rate, positioning error and system interference as the optimization direction, take the coordinates and excitation phase of each reader antenna as the optimization variables, and establish the optimization model. in represents the candidate solution of the RNP problem, (x m ,y m ) represents the coordinates of the reader antenna, α m represents the excitation phase difference of the reader antenna, m∈[1,M], M is the number of reader antennas, F=ω1f1+ω2f2+ω3f3, represents the RNP problem optimization objective function, ω1, ω2 and ω3 represent the inertia weights of the three sub-functions f1, f2 and f3 respectively, It is used to measure the non-coverage of labels, where n∈[1,N], N represents the total number of labels, and Cn The nth target tag can only be successfully located when it is recognized by three or more readers. Used to measure positioning error, GDOP n is the geometric precision factor of the nth label, f3 = k T I T +k R I R , used to measure the interference between the tag and the reader, where I T It represents the average value of the interference degree of all tags by the reader, I R represents the average value of the interference degree of all readers by tags, k T and k R Indicates the degree of reader interference and tag interference;

[0010] Step 5: Design an improved sparrow search algorithm to solve the RNP problem established in step 4. First, establish the initial solution Ω according to the initial position of the reader antenna. I , and then randomly perturb Ω within the preset optimization radius I , thereby generating an initial population P of size Q 1 , further evaluate the fitness value of each individual, divide the population into discoverers, joiners and alerters, and finally calculate the distance d from each individual to other individuals and calculate the evolution factor f F , calculate the membership function μ(f), determine the current evolutionary stage according to the membership function, and use the corresponding strategy for iterative update. When μ(f)∈S1, the population is in the exploration stage, and the traditional sparrow search algorithm is used for update. When μ(f)∈S2, the population is in the convergence stage, and the orthogonal experimental design OED (Orthogonal Experimental Design) update strategy is used for update. When μ(f)∈S3, the population is in the convergence stage, and the generative adversarial network GAN (Generative Adversarial Network) update strategy is used for update.

[0011] Step 6: Perform iterative operations on the population. When the number of iterative operations reaches the upper limit, terminate the optimization process and output the final optimal deployment method. According to this method, the RFID antenna optimization deployment method based on single-base communication and phased array is completed.

[0012] It should be noted that the improved sparrow search algorithm in step 5 combines two update strategies and an update method selection mechanism. The proposed algorithm can evaluate the evolution process based on the distance between sparrow individuals, classify the evolution process into one of the stages S1 (exploration), S2 (convergence) and S3 (jump out), and finally select the corresponding method to update the population according to the evolution stage. The details are as follows:

[0013] The process of updating the strategy with OED in step 5 is, first, to i Domain i ,u i ]Quantized into Q1 levels Where Q1 is an odd number, the difference between any two consecutive levels is the same, α i，j represents the jth level in the i-th dimension. After the domain of the solution is quantized, x i There are Q1 possible values Therefore, Q1 will be evenly distributed in the D-dimensional feasible solution space. D However, since the problem dimension D is given, there may not be Q1 and J1 that satisfy So choose the smallest J1 to satisfy Then, construct the basic matrix and the non-basic matrix, merge the basic matrix and the non-basic matrix, and let all a i,j =a i，j +1, generate an orthogonal matrix where a i，j Represents the element in the i-th row and j-th column of the orthogonal matrix. Since D′ is usually greater than or equal to D, the matrix needs to be The last D′-D columns of are deleted to obtain the final orthogonal matrix in Further, in constructing After the matrix, we can get D Then, based on the M1 combination samples, a population containing M1 individuals can be generated. Finally, for the population Conduct a fitness value evaluation and select the N individuals with the best fitness values ​​to generate a new generation of sparrow population;

[0014] The process of using GAN to update the strategy in step 5 is, first, initialize a pair of GANs, define the generator and discriminator networks by setting initial parameters, classify the candidate solutions in the population P into two different sample sets of equal size, true and false, and use them to train GAN. Then, perform multiple iterations of training on GAN. In each iteration, three different types of training samples are used to update the generator and discriminator, namely true samples, false samples, and samples generated by the generator. The loss function of training the discriminator is expressed as Where G represents the generator, D represents the discriminator, r represents the true sample, f represents the fake sample, z represents the random input data, which is random Gaussian noise, D(r), D(f) and D(G(z)) represent the output of the true sample, the fake sample and the sample generated by the generator input to the discriminator respectively. In each GAN training iteration, all samples in X are randomly divided into several batches of size m. According to the gradient descent method, the samples generated by itself are used to update the discriminator and generator. The update formula is expressed as where z i is a vector randomly sampled from a Gaussian distribution. Further, the D-dimensional vector x is sampled, where each element is independently sampled from a continuous uniform distribution U(0, 1). Based on the probability density function, a D-dimensional vector y that satisfies the multivariate normal distribution is generated, which is expressed as Where D represents the dimension of the decision space, ∑ and μ represent the covariance matrix and mean vector of the true sample, respectively, expressed as where r i Represents the i-th member of the true sample set, N represents the population size, and finally, the output of the generator G(y) is the final solution x′=G(y)(ul)+l, where x′ represents the candidate solution generated by GAN, u represents the upper bound of the solution space, and l represents the lower bound of the solution space.

[0015] The present invention provides an RFID antenna optimization deployment method based on single-base communication and phased array. The method first derives the gain expression of a one-dimensional linear phased array antenna based on the gain expression of a two-dimensional phased array antenna and the principle of calculus. Then, based on the expression, a channel link model based on a one-dimensional phased array antenna is established to simulate the signal propagation state. Furthermore, an RNP problem optimization model is constructed by combining tag coverage, positioning accuracy and system interference. It has certain guiding significance for improving the quality and efficiency of solving the RNP problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a system framework diagram of the present invention;

[0017] Figure 2 is a schematic diagram of the radiation scene under discrete conditions of the present invention;

[0018] Figure 3 It is a schematic diagram of the structure of the linear phased array antenna of the present invention;

[0019] Figure 4 is a schematic diagram of a Cartesian coordinate system under the integration conditions of the present invention;

[0020] Figure 5 It is a schematic diagram of the flow chart of the improved sparrow search algorithm of the present invention;

[0021] Figure 6It is a schematic diagram of the orthogonal experimental design (OED) strategy flow chart of the present invention;

[0022] Figure 7 It is a schematic diagram of the model training scheme of GAN of the present invention;

[0023] Figure 8 It is a schematic diagram of the fuzzy membership function diagram of three evolutionary states of the present invention; DETAILED DESCRIPTION

[0024] The purpose of the present invention is to provide an RFID antenna optimization deployment method based on single-base communication and phased array. The method first derives the gain expression of the one-dimensional linear phased array antenna based on the gain expression of the two-dimensional phased array antenna and the principle of calculus. Then, based on the expression, a channel link model based on the one-dimensional phased array antenna is established to simulate the signal propagation state. Further, an RNP problem optimization model is constructed in combination with tag coverage, positioning accuracy and system interference. It has certain guiding significance for improving the quality and efficiency of solving the RNP problem.

[0025] The following is combined with Figure 1 , Attachment Figure 2 , Attachment Figure 3 , Attachment Figure 4 , Attachment Figure 5 , Attachment Figure 6 , Attachment Figure 7 And attached Figure 8 The embodiments of the present invention are described in further detail.

[0026] The present invention is based on the design concept of single-base communication and phased array and is designed based on the characteristics and requirements of RFID systems in practical applications.

[0027] The system framework of the present invention is shown in the attached Figure 1 As shown in the figure, the system framework mainly includes: establishing a channel link model based on linear phased array antenna to accurately estimate the received power in the forward link and reverse link between the reader and the tag; using a half-wave dipole antenna as the tag antenna, and deriving the gain expression of the one-dimensional linear phased array antenna based on the gain expression of the two-dimensional phased array antenna; integrating the gain models of the reader antenna and the tag into the same Cartesian coordinate system; building a channel link model based on this, and establishing an optimization model for the RNP problem in the single-base communication mode in combination with the optimization objectives such as tag coverage, positioning accuracy and system interference; proposing an improved sparrow search algorithm EGOSSA to solve the RNP problem; performing iterative operations on the population, and when the number of iterative operations reaches the upper limit, terminating the optimization process and outputting the final optimal deployment method, according to which the RFID antenna optimization deployment method based on single-base communication and phased array is completed.

[0028] As attached Figure 2As shown in the figure, in practical applications, RFID systems usually use half-wavelength dipole antennas as tag antennas, and establish Figure 2 The radiation scenario shown, its gain can be expressed as G T (θ T ,φ T )=1.641(cos 2 (0.5πcos(θ T ))) / sin 2 (θ T ), where the long side of the tag antenna is the z-axis and the short side is the x-axis, is the radiation direction, θ T and φ T represents a pair of directivity parameters, θ T express arrive The inclination angle, φ T express arrive The rotation angle of the projection on the XOY plane.

[0029] A one-dimensional linear phased array antenna is used as the reader antenna, as shown in the attached Figure 3 As shown, its gain can be expressed as G R (θ R ,φ R )=eD(θ R ,φ R ),in e represents the efficiency factor, D(θ R ,φ R ) represents the directivity coefficient, S(θ R ,φ R ) represents the planar array factor, I n They represent the excitation amplitude along the y-axis, n represents the array element index, and y n represents the coordinate of the element along the y-axis, α represents the phase difference of the excitation, and θ R and φ R Definition and θ T and φ T Similarly, let cosθ y = sinθ R cosφ R ,S(θ R ,φ R ) can be expressed as Among them, θ y Indicates the y-axis and The angle between the two, N represents the number of linear array elements, I n represents the excitation amplitude along the y-axis, y nrepresents the coordinate of the array element along the y-axis, α is the phase difference of the excitation, and a 1×4 linear phased array antenna is used. Assuming e=1, I n =1, k = 2π / λ, d = 0.5λ, the array element coordinates are (0, -0.75λ, 0), (0, -0.25λ, 0), (0, 0.25λ, 0), (0, 0.75λ, 0), the planar array factor S(θ y ) can be expressed as After formula derivation and definite integral calculation, the gain of the reader antenna can be expressed as

[0030] Considering the directivity of the gain of the reader antenna and the tag, the gain models of the reader antenna and the tag are integrated into the same Cartesian coordinate system, as shown in the attached figure. Figure 4 As shown. (x T ,y T , z T ) and (x R ,y R , z R ) represent the spatial coordinates of the tag antenna and the reader antenna, and They represent the positions of the reader antenna and the tag antenna respectively, where Indicates from arrive The inclination angle, Indicates from arrive The rotation angle mapped in the XOY plane, Representative from arrive The inclination angle, express arrive In the XOY plane mapping rotation angle, the reader antenna is usually located higher than the tag antenna and radiates at a downward angle, G T and G R Can be used and Represented as G T =1.641cos 2 (A T / B T ) / (1-(C T / D T ) 2 ), in

[0031] Based on this, a channel link model is established, and the RNP problem optimization model under the single-base communication mode is established by combining the optimization objectives such as tag coverage, positioning accuracy and system interference.

[0032] As attached Figure 5 As shown in the figure, an improved sparrow search algorithm is designed to solve the established RNP problem. The proposed algorithm can evaluate the evolution process based on the distance between individual sparrows, classify the evolution process into one of the stages S1 (exploration), S2 (convergence) and S3 (jump out), and finally select the corresponding method to update the population according to the evolution stage. First, the initial solution Ω is established according to the initial position of the reader antenna. I , and then randomly perturb Ω within the preset optimization radius I , thereby generating an initial population P of size Q 1 , further evaluate the fitness value of each individual, divide the population into discoverers, joiners and alerters, and finally calculate the distance d from each individual to other individuals and calculate the evolution factor f F , calculate the membership function μ(f), and judge the current evolution stage according to the membership function. This paper divides the search into three stages: exploration, convergence and jump out. The three stages are assigned as shown in the following Figure 8 The fuzzy membership function shown in the figure is iteratively updated using the corresponding strategy. When μ(f)∈S1, the population is in the exploration stage and the update strategy of the traditional sparrow search algorithm is used for updating. When μ(f)∈S2, the population is in the convergence stage and the update strategy of OED (Orthogonal Experimental Design) is used for updating. When μ(f)∈S3, the population is in the convergence stage and the update strategy of GAN (Generative Adversarial Network) is used for updating.

[0033] As attached Figure 6 As shown in step 5, when the OED update strategy is used to update the population, first, x i Domain i ,u i ]Quantized into Q1 levels Where Q1 is an odd number, the difference between any two consecutive levels is the same, α i，j represents the jth level in the i-th dimension. After the domain of the solution is quantized, x i There are Q1 possible values Therefore, Q1 will be evenly distributed in the D-dimensional feasible solution space. D However, since the problem dimension D is given, there may not be Q1 and J1 that satisfy So choose the smallest J1 to satisfy Then, construct the basic matrix and the non-basic matrix, merge the basic matrix and the non-basic matrix, and let all a i,j =a i,j +1, generate an orthogonal matrix where a i，j Represents the element in the i-th row and j-th column of the orthogonal matrix. Since D′ is usually greater than or equal to D, the matrix needs to be The last D′-D columns of are deleted to obtain the final orthogonal matrix in Further, in constructing After the matrix, we can get D Then, based on the M1 combination samples, a population containing M1 individuals can be generated. Finally, for the population Perform a fitness value evaluation and select the N individuals with the best fitness values ​​to generate a new generation of sparrow population.

[0034] As attached Figure 7 As shown in Figure 5, when the GAN update strategy is used to update the population in step 5, first, a pair of GANs is initialized. By setting the initial parameters, the generator and discriminator networks are defined, and the candidate solutions in the population P are classified into two different sample sets of equal size, true and false, and used to train the GAN. Then, the GAN is trained for multiple iterations. In each iteration, three different types of training samples are used to update the generator and discriminator, namely, true samples, false samples, and samples generated by the generator. The loss function of training the discriminator is expressed as Where G represents the generator, D represents the discriminator, r represents the true sample, f represents the fake sample, z represents the random input data, which is random Gaussian noise, D(r), D(f) and D(G(z)) represent the output of the true sample, the fake sample and the sample generated by the generator input to the discriminator respectively. In each GAN training iteration, all samples in X are randomly divided into several batches of size m. According to the gradient descent method, the samples generated by itself are used to update the discriminator and generator. The update formula is expressed as where z i is a vector randomly sampled from a Gaussian distribution. Further, the D-dimensional vector x is sampled, where each element is independently sampled from a continuous uniform distribution U(0, 1). Based on the probability density function, a D-dimensional vector y that satisfies the multivariate normal distribution is generated, which is expressed as Where D represents the dimension of the decision space, ∑ and μ represent the covariance matrix and mean vector of the true sample, respectively, expressed as where r iRepresents the i-th member of the true sample set, N represents the population size, and finally, the output of the generator G(y) is the final solution x′=G(y)(ul)+l, where x′ represents the candidate solution generated by GAN, u represents the upper bound of the solution space, and l represents the lower bound of the solution space.

Claims

1. A method for optimizing the deployment of RFID antennas based on single-base communication and phased array, characterized in that: The specific steps are as follows: Step 1: Establish a channel link model based on a linear phased array antenna to accurately estimate the received signal power in the forward link and reverse link between the reader and the tag. The Radio Frequency Identification (RFID) system usually adopts an RFID single base station system. The received signal power P of the tag in the forward link is r，T =10log 10 (ρ L P Tx G T G T L(d))+30, the received signal power of the reader antenna in the backward link where ρ L represents the polarization loss factor, P Tx Indicates the transmission power of the reader, G T and G R They represent the gains of the tag antenna and the reader antenna, μ T represents the power transmission efficiency, κ represents the modulation efficiency, Γ is the differential reflection coefficient of the tag, and L(d) represents the path loss factor, which is specifically expressed as L(d) = λ 2 / (16π 2 η 2 d 2 ), where d represents the distance between the reader and the tag, and η represents the path loss exponent; Step 2: Use a half-wave dipole antenna as the tag antenna, and its gain can be expressed as G T (θ T ,φ T )=1.641(cos 2 (0.5πcos(θ T ))) / sin 2 (θ T ), where the long side of the tag antenna is the z-axis and the short side is the x-axis, is the radiation direction, θ T and φ T represents a pair of directivity parameters, θ T express arrive The inclination angle, φ T express arrive The rotation angle projected on the XOY plane uses a one-dimensional linear phased array antenna as the reader antenna, and its gain can be expressed as G R (θ R ,φ R )=eD(θ R ,φ R ),in e represents the efficiency factor, D(θ R ,φ R ) represents the directivity coefficient, S(θ R ,φ R ) represents the plane array factor, I n They represent the excitation amplitude along the y-axis, n represents the array element index, and y n represents the coordinate of the element along the y-axis, α represents the phase difference of the excitation, and θ R and φ R Definition and θ T and φ T Similarly, let cosθ y = sinθ R cosφ R ,S(θ R ,φ R ) can be expressed as Among them, θ y Indicates the y-axis and The angle between the two, N represents the number of linear array elements, I n represents the excitation amplitude along the y-axis, y n represents the coordinate of the array element along the y-axis, α is the phase difference of the excitation, and a 1×4 linear phased array antenna is used. Assume e=1, I n =1, k = 2π / λ, d = 0.5λ, the array element coordinates are (0, -0.75λ, 0), (0, -0.25λ, 0), (0, 0.25λ, 0), (0, 0.75λ, 0), the planar array factor S(θ y ) can be expressed as After formula derivation and definite integral calculation, the gain of the reader antenna can be expressed as Step 3: Integrate the gain models of the reader antenna and the tag into the same Cartesian coordinate system. (x T ,y T , z T ) and (x R ,y R , z R ) represent the spatial coordinates of the tag antenna and the reader antenna, and They represent the positions of the reader antenna and the tag antenna respectively, where Indicates from arrive The inclination angle, Indicates from arrive The rotation angle mapped in the XOY plane, Representative from arrive The inclination angle, express arrive In the XOY plane mapping rotation angle, the reader antenna is usually located higher than the tag antenna and radiates at a downward angle, G T and G R Can be used and Represented as G T =1.641cos 2 (A T / B T ) / (1-(C T / D T ) 2 ), in Step 4: Establish the RNP (RFID Network Planning) problem model, take the minimization of coverage rate, positioning error and system interference as the optimization direction, take the coordinates and excitation phase of each reader antenna as the optimization variables, and establish the optimization model. Where Ω=[(x1, y1, α1), (x2, y2, α2),..., (x M ,y M , α M )], represents the candidate solution of the RNP problem, (x m ,y m ) represents the coordinates of the reader antenna, α m represents the excitation phase difference of the reader antenna, m∈[1,M], M is the number of reader antennas, F=ω1f1+ω2f2+ω3f3, represents the RNP problem optimization objective function, ω1, ω2 and ω3 represent the inertia weights of the three sub-functions f1, f2 and f3 respectively, It is used to measure the non-coverage of labels, where n∈[1,N], N represents the total number of labels, and C n The nth target tag can only be successfully located when it is recognized by three or more readers. Used to measure positioning error, GDOP n is the geometric precision factor of the nth label, f3 = k T I T +k R I R , used to measure the interference between the tag and the reader, where I T It represents the average value of the interference degree of all tags by the reader, I R represents the average value of the interference degree of all readers by tags, k T and k R Indicates the degree of reader interference and tag interference; Step 5: Design an improved sparrow search algorithm to solve the RNP problem established in step 4. First, establish the initial solution Ω according to the initial position of the reader antenna. I , and then randomly perturb Ω within the preset optimization radius I , thereby generating an initial population P of size Q 1 , further evaluate the fitness value of each individual, divide the population into discoverers, joiners and alerters, and finally calculate the distance d from each individual to other individuals and calculate the evolution factor f F , calculate the membership function μ(f), determine the current evolutionary stage according to the membership function, and use the corresponding strategy for iterative update. When μ(f)∈S1, the population is in the exploration stage, and the traditional sparrow search algorithm is used for update. When μ(f)∈S2, the population is in the convergence stage, and the orthogonal experimental design OED (Orthogonal Experimental Design) update strategy is used for update. When μ(f)∈S3, the population is in the convergence stage, and the generative adversarial network GAN (Generative Adversarial Network) update strategy is used for update. Step 6: Perform iterative operations on the population. When the number of iterative operations reaches the upper limit, terminate the optimization process and output the final optimal deployment method. According to this method, the RFID antenna optimization deployment method based on single-base communication and phased array is completed.

2. The method according to claim 1, characterized in that: The process of updating the strategy with OED in step 5 is, first, to i Domain i ,u i ]Quantized into Q1 levels Where Q1 is an odd number, the difference between any two consecutive levels is the same, α i，j represents the jth level in the i-th dimension. After the domain of the solution is quantized, x i There are Q1 possible values Therefore, Q1 will be evenly distributed in the D-dimensional feasible solution space. D However, since the problem dimension D is given, there may not be Q1 and J1 that satisfy So choose the smallest J1 to satisfy Then, construct the basic matrix and the non-basic matrix, merge the basic matrix and the non-basic matrix, and let all a i,j =a i，j +1, generate an orthogonal matrix where a i，j Represents the element in the i-th row and j-th column of the orthogonal matrix. Since D′ is usually greater than or equal to D, the matrix needs to be The last D′-D columns of are deleted to obtain the final orthogonal matrix in Further, in constructing After the matrix, we can get D Then, based on the M1 combination samples, a population containing M1 individuals can be generated. Finally, for the population Perform a fitness value evaluation and select the N individuals with the best fitness values ​​to generate a new generation of sparrow population.

3. The method according to claim 1, characterized in that In step 5, the process of the GAN update strategy used is, first, initialize a pair of GANs, define the generator and discriminator networks by setting initial parameters, classify the candidate solutions in the population P into two different sample sets of equal size, true and false, and use them to train the GAN, then perform multiple iterations of training on the GAN, in each iteration, use three different types of training samples to update the generator and discriminator, i.e., true samples, false samples, and samples generated by the generator, and the loss function of training the discriminator is expressed as Where G represents the generator, D represents the discriminator, r represents the true sample, f represents the fake sample, z represents the random input data, which is random Gaussian noise, D(r), D(f) and D(G(z)) represent the output of the true sample, the fake sample and the sample generated by the generator input to the discriminator respectively. In each GAN training iteration, all samples in X are randomly divided into several batches of size m. According to the gradient descent method, the samples generated by itself are used to update the discriminator and generator. The update formula is expressed as where z i is a vector randomly sampled from a Gaussian distribution. Further, the D-dimensional vector x is sampled, where each element is independently sampled from a continuous uniform distribution U(0, 1). Based on the probability density function, a D-dimensional vector y that satisfies the multivariate normal distribution is generated, which is expressed as Where D represents the dimension of the decision space, ∑ and μ represent the covariance matrix and mean vector of the true sample, respectively, expressed as where r i Represents the i-th member of the true sample set, N represents the population size, and finally, the output of the generator G(y) is the final solution x′=G(y)(ul)+l, where x′ represents the candidate solution generated by GAN, u represents the upper bound of the solution space, and l represents the lower bound of the solution space.

Images