5G mine underground positioning method of consistency specificity multi-view subspace clustering algorithm based on convex relaxation constraint

Abstract

The invention belongs to the technical field of 5G mine underground positioning, and particularly relates to a 5G mine underground positioning method based on a convex relaxation constraint consistency specificity multi-view subspace clustering algorithm. Firstly, in a sampling part, sampling point position information and multiple signal intensities of a 5G access point are obtained, and high-dimensional data points are combined; in the clustering stage, the data are combined into a matrix, clustering is conducted through a non-negative matrix factorization multi-view clustering algorithm, and a clustering result is obtained through a specific optimization process and convex relaxation constraint processing; in the positioning stage, the position of the to-be-measured point is estimated according to the RSSI. According to the method, the problems that a traditional positioning method only uses a single signal parameter and a multi-view clustering algorithm does not consider data characteristics are solved, multiple signal parameters are comprehensively utilized, the algorithm is optimized, the positioning precision is improved, experiments prove that the positioning success rate of the method is higher than that of a traditional algorithm, the method is suitable for 5G mine personnel positioning, and powerful support is provided for mine safety production.

Description

Technical Field

[0001] The present invention belongs to the technical field of 5G underground mine positioning, and in particular relates to a 5G underground mine positioning method based on a consistency-specific multi-view subspace clustering algorithm with convex relaxation constraints. Background Art

[0002] In order to improve the positioning accuracy of personnel in mines, a 5G underground mine positioning method based on a convex relaxation constraint consistency-specific multi-view subspace clustering algorithm is proposed. This method saves the three signal strength parameters and location information at the sampling point to form a database of three views, and uses a convex relaxation constraint consistency-specific multi-view subspace clustering algorithm to cluster the database to obtain the label information of each category. The 5G terminal compares the AP signal strength measured in real time with the cluster label information in the database, and calculates the terminal's position using the Euclidean distance formula.

[0003] The traditional underground mine positioning method based on clustering algorithm uses RSSI (Received Signal Strength Indication) as the parameter of signal strength when constructing the database. Then the location information and signal strength of the 5G wireless access point (AP) are combined to obtain high-dimensional data points, and then the non-negative matrix decomposition clustering algorithm is applied to it to finally obtain the clustering result. However, in this process, other information of the 5G signal is ignored. In 5G communication, in addition to RSSI, the commonly used signal strength parameters include RSRP (Reference Signal Receiving Power) and RSRQ (Reference Signal Receiving Quality). The current underground mine positioning method based on clustering algorithm ignores these two parts of information, which may result in a large error value in the positioning result. As a result, the exact location of the target point cannot be accurately obtained, thereby affecting the accuracy of the entire underground personnel positioning system. This algorithm proposes: when constructing the database of sampling points, three data matrices based on the three parameters of RSSI, RSRP, and RSRQ are constructed at the same time, and then the database composed of these three data matrices is subjected to the consistent specific multi-view subspace clustering algorithm based on convex relaxation constraints, so that the clustering labels obtained will be more reliable.

[0004] Although the traditional underground mine positioning method based on multi-view clustering algorithm takes different signal strength parameters into consideration when constructing the database, this method simply adds the three signal strength parameter matrices during clustering, and only considers the consistency of the data, but not the diversity of the data. Therefore, this algorithm proposes a consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints. This method formulates multi-view self-representation properties by sharing a consistent representation and a set of specific representations. The resulting clustering results will be more accurate because the consistency and specificity of the three signal strength parameters are taken into account. Summary of the invention

[0005] Aiming at the technical problems existing in the background technology, the present invention introduces convex relaxation and develops an alternating optimization algorithm to restore the corresponding data representation in order to optimize the non-convex problem on the basis of the first two points, so as to achieve better clustering effect. A 5G underground mine positioning method based on a convex relaxation-constrained consistency-specific multi-view subspace clustering algorithm is proposed.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is: comprising the following steps:

[0007] S1. First, the sampling part obtains the location information of n sampling points. Each sampling point collects the signal strength of d 5G access points, including signal strength indication RSSI, signal received power RSRP and signal received quality RSRQ. The location information of each sampling point is combined with the three signal strength parameter information into three d+2-dimensional high-dimensional data points;

[0008] S2, followed by the clustering stage, the three groups of n vectors with dimension d+2 are combined to obtain three matrices with dimension (d+2)×n, and the three matrices are clustered using the non-negative matrix decomposition multi-view clustering algorithm to obtain a set of k categories C = {c1, c2, ..., c k};

[0009] S3, the last is the positioning stage, by detecting the RSSI of the point to be measured, the position of the point to be measured is estimated;

[0010] The implementation process of the non-negative matrix decomposition multi-view clustering algorithm in step S2 is:

[0011] S21, first define a data matrix X with a dimension of d×n = {x1, x2, ...x n}, where d is the dimension of each data point and n is the number of sample data points;

[0012] S22. For each data point x i, belongs to one of C categories, and reconstructs itself through an n×n self-representation matrix Z. For multiple views: X v =X v Z+E v , where Z is the self-representation matrix, X v , E v Represents the data and error of the vth view, and all views share a self-representation matrix Z;

[0013] S23. Optimize the above formula to consistent specific multi-view subspace clustering: X v =X v (C+D v )+E v , where C, D v Represent the consistent and specific self-representation matrix learned for the data under different views, that is, decompose the matrix Z into the sum of shared representations, which reconstructs the view-specific representation of the part that remains unchanged across views and the unique part corresponding to the v-th view;

[0014] S24, using the consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints for processing, selecting the nuclear norm for consistent items to ensure low-rank properties, and applying the L2 norm for specific items to ensure connectivity: xT v =X v (C+D v )+E v Among them, || || * represents the nuclear norm of the matrix, represents the square of the matrix's bi-norm, || || 2,1 represents the 2,1 norm of the matrix, λ C ,λ D It is a parameter set to balance the regularization of the objective function;

[0015] S25, finally get C and D v Then, construct the similarity matrix S:

[0016] Perform spectral clustering on the matrix S to obtain the final clustering result.

[0017] 5. According to a 5G underground mine positioning method based on a convex relaxation constraint consistency specific multi-view subspace clustering algorithm according to claim 1, it is characterized in that the specific operation of combining the position information of each sampling point and the three signal strength parameter information into three d+2-dimensional high-dimensional data points in step S1 is:

[0018] S11. First, define the location information of n sampling points as (x1, y1), (x2, y2), (x3, y3)..., (x n ,y n ), where n is the number of sampling points;

[0019] S12. Each sampling point can obtain the signal strength of d 5GAPs, where the signal strength indication RSSI of the i-th sampling point is expressed as The signal received power RSRP is expressed as The signal reception quality RSRQ is expressed as

[0020] S13. Finally, the position signal of each sampling point is combined with their three signal strength parameter information RSSI, RSRP, and RSRQ to obtain three d+2-dimensional high-dimensional data points, which are respectively expressed as:

[0021]

[0022] 6. According to the 5G underground mine positioning method of the consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints in claim 1, it is characterized in that the specific solution process of the consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints in step S2 is: using convex relaxation and introducing intermediate variables to perform alternating optimization solution, first introducing intermediate variables K and W v The formula in step S24 is converted into: xT v =X v (C+D v )+E v , C=K,E v =W v ; Then solve it by the method of augmented Lagrange multipliers:

[0023]

[0024] where <> is the standard Euclidean inner product of two matrices, || || F represents the norm of the matrix, is the Lagrange multiplier, and μ is the penalty term coefficient.

[0025] 7. According to the 5G underground mine positioning method based on the convex relaxation constraint consistency specific multi-view subspace clustering algorithm of claim 1, it is characterized in that the specific implementation of step S3 is:

[0026] S31. First, for each data point in each category of the collected samples, remove the position information (x i ,yi ) Only keep RSSI data i = {r i1 ,r i2 ,...,r id};

[0027] S32, then for each category, find the mean signal strength of the data points in these categories to obtain k mean points r k = {r k1 ,r k2 ,...,r kd};

[0028] S33, then calculate the Euclidean distance between the k mean points and the signal strength vector of the detected signal point, and obtain the category corresponding to the minimum value;

[0029] S34. Finally, the position information of all vectors in this category is averaged to estimate the position of the detection point.

[0030] Compared with the prior art, the advantages and positive effects of the present invention are that it is more comprehensive in data utilization, taking into account the three signal strength parameters of RSSI, RSRP and RSRQ, making the clustering labels more reliable. The adopted consistency-specific multi-view subspace clustering algorithm takes into account both data consistency and diversity, and the clustering results are more accurate. By introducing convex relaxation and alternating optimization algorithms, non-convex problems are optimized and the clustering effect is improved. Experiments show that the positioning success rate of the present invention is higher. For example, in underground experiments in coal mines, the positioning success rate is significantly improved compared with traditional algorithms, providing a more accurate and effective solution for the positioning of mine personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 Schematic diagram of the positioning structure for underground personnel;

[0033] Figure 2 This is a schematic diagram of the positioning principle of underground personnel. DETAILED DESCRIPTION

[0034] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0035] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0036] Example 1. In a mine operating environment, accurate personnel positioning is crucial to ensuring operational safety and improving production efficiency. Traditional positioning methods only use a single signal strength parameter such as RSSI when constructing a database, ignoring important information such as RSRP and RSRQ, resulting in large positioning errors. Although some multi-view clustering algorithms consider multiple parameters, the clustering process is simple and does not take into account data consistency and diversity. Based on this, the present invention proposes a 5G underground mine positioning method based on a consistency-specific multi-view subspace clustering algorithm with convex relaxation constraints. The system structure of the present invention consists of a ground control center and an underground 5GAP (access point) and a 5G mobile terminal carried by the operator, such as Figure 1 As shown in the figure, the 5G underground mine positioning method based on non-negative matrix decomposition multi-view clustering mainly consists of two stages: training stage and real-time positioning stage. Figure 2 shown.

[0037] The first is the training phase, which consists of the sampling phase and the clustering phase. For the sampling part, the location information of n sampling points is obtained. Each sampling point collects the signal strength of d 5G access points, including signal strength indication RSSI, signal received power RSRP and signal received quality RSRQ. The location information of each sampling point is combined with the three signal strength parameter information into three d+2-dimensional high-dimensional data points. The realization of three d+2-dimensional high-dimensional data points is to first define the location information of n sampling points as (x1, y1), (x2, y2), (x3, y3)..., (x n ,y n ), where n is the number of sampling points; each sampling point can obtain the signal strength of d 5GAPs, where the signal strength indication RSSI of the i-th sampling point is expressed as The signal received power RSRP is expressed as The signal reception quality RSRQ is expressed as Finally, the position signal of each sampling point is combined with their three signal strength parameter information RSSI, RSRP, and RSRQ to obtain three d+2-dimensional high-dimensional data points, which are expressed as:

[0038] In the clustering stage, three groups of n vectors with dimension d+2 are obtained in the above operation. By combining them, three matrices with dimension (d+2)×n can be obtained. The non-negative matrix decomposition multi-view clustering algorithm is used for these three matrices. First, define the data matrix X={x1,x2,...x n}, where d is the dimension of each data point and n is the number of sample data points; for each data point x i , belongs to one of C categories, and reconstructs itself through an n×n self-representation matrix Z. For multiple views: X v =X v Z+E v , where Z is the self-representation matrix, X v , E v Represents the data and error of the vth view, and all views share a self-representation matrix Z; the above formula is optimized to the consistency-specific multi-view subspace clustering: X v =X v (C+D v )+E v , where C, D v Represent the consistent and specific self-representation matrix learned from the data under different views, that is, decompose the matrix Z into the sum of shared representations, which reconstructs the view-specific representation of the parts that remain unchanged on different views and the unique part corresponding to the v-th view; the consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints is used for processing, and the nuclear norm is selected for consistent items to ensure low-rank properties, and the L2 norm is applied to specific items to ensure connectivity: xT v =X v (C+D v )+E v Among them, || || * represents the nuclear norm of the matrix, represents the square of the matrix's bi-norm, || || 2,1 represents the 2,1 norm of the matrix, λ C ,λ D It is a parameter set to balance the regularization of the objective function. Use convex relaxation and introduce intermediate variables to solve C and D in alternating optimization v , first introduce the intermediate variables K and W v Will xT v =X v (C+D v )+E v The formula is transformed into: xT v =X v (C+Dv )+E v , C=K,E v =W v ; Then solve it by the method of augmented Lagrange multipliers:

[0039]

[0040] where <> is the standard Euclidean inner product of two matrices, || || F represents the norm of the matrix, is the Lagrange multiplier, μ is the penalty term coefficient. v Then, construct the similarity matrix S: Perform spectral clustering on the matrix S to obtain the final clustering result.

[0041] The specific process of this algorithm is as follows: first, step 1, initialization parameters, where C = 0, K = 0, μ=10 -6 , Step 2: Fix other variables according to Formula updates C, and then step 3, fix other variables, according to Update K, step 4, fix other variables, according to Update D v , Step 5, fix other variables, according to Update E v , Step 6, fix other variables, according to Update W v , Step 7, fix other variables, according to Y 2,t+1 =Y 2,t +μ(CK), Finally, repeat steps 2 to 7 until C, D v Convergence, according to Calculate the similarity matrix S, and cluster S according to traditional spectral clustering to output the category set C = {c1, c2, ...c k}.

[0042] Finally, in the positioning stage, the position of the point to be measured is estimated by detecting the RSSI of the point to be measured. First, for each data point in each category of the collected samples, the position information (x i ,y i ) Only keep RSSI data i = {r i1 ,r i2 ,...,r id}; Then for each category, find the mean signal strength of the data points in these categories and get k mean points r k = {rk1 ,r k2 ,...,r kd}; Then calculate the Euclidean distance between the signal strength vector of these k mean points and the detected signal point, and get the category corresponding to the minimum value; Finally, calculate the average of the position information part of all vectors in this category to estimate the position of the detection point.

[0043] In Example 2, an underground tunnel of a coal mine in Huangling was selected for the experiment. According to the environment and direction of the tunnel, the distance between 5GAPs was controlled within 90m. A 5G mine intrinsically safe wireless communication system was selected as the 5GAP, and a mine 5G intrinsically safe mobile phone was used as the mobile terminal for the operator. After explosion-proof testing, all the equipment used met the requirements for the intrinsic safety of underground coal mines.

[0044] During the experiment, a total of 1200 sampling points were set up, and each sampling point collected RSSI, RSRQ, and RSRQ data 10 times. After obtaining the sampling database, the testers conducted 100 tests as test data.

[0045] The sampling point data is stored in the original data server. In the experiment, we set the number of categories k from 1 to 9, and then use the non-negative matrix factorization clustering algorithm and the non-negative matrix factorization multi-view clustering algorithm respectively. The positioning success rate (error less than 2m) is shown in Table 1. The experimental results show that no matter what value k takes, the positioning success rate of the non-negative matrix factorization clustering algorithm is much higher than that of the k nearest neighbor algorithm. At the same time, when k is 3, the success rate of the algorithm reaches the maximum value.

[0046] Table 1: Comparison of the localization success rates of the k-nearest neighbor algorithm and the convex relaxation constraint consistency-specific multi-view subspace clustering algorithm

[0047]

[0048]

[0049] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A 5G underground mine positioning method based on a convex relaxation constraint-based consistency-specific multi-view subspace clustering algorithm, characterized in that: The following steps are involved: S1. First, the sampling part obtains the location information of n sampling points. Each sampling point collects the signal strength of d 5G access points, including signal strength indication RSSI, signal received power RSRP and signal received quality RSRQ. The location information of each sampling point is combined with the three signal strength parameter information into three d+2-dimensional high-dimensional data points; S2, followed by the clustering stage, the three groups of n vectors with dimension d+2 are combined to obtain three matrices with dimension (d+2)×n, and the three matrices are clustered using the non-negative matrix decomposition multi-view clustering algorithm to obtain a set of k categories C = {c1, c2, ..., c k }; S3, the last is the positioning stage, by detecting the RSSI of the point to be measured, the position of the point to be measured is estimated; The implementation process of the non-negative matrix decomposition multi-view clustering algorithm in step S2 is: S21, first define a data matrix X with a dimension of d×n = {x1, x2, ...x n }, where d is the dimension of each data point and n is the number of sample data points; S22. For each data point x i , belongs to one of C categories, and reconstructs itself through an n×n self-representation matrix Z. For multiple views: X v =X v Z+E v , where Z is the self-representation matrix, X v , E v Represents the data and error of the vth view, and all views share a self-representation matrix Z; S23. Optimize the above formula to consistent specific multi-view subspace clustering: X v =X v (C+D v )+E v , where C, D v Represent the consistent and specific self-representation matrix learned for the data under different views, that is, decompose the matrix Z into the sum of shared representations, which reconstructs the view-specific representation of the part that remains unchanged across views and the unique part corresponding to the v-th view; S24, using the consistency-specific multi-view subspace clustering algorithm based on convex relaxation constraints for processing, selecting the nuclear norm for consistent items to ensure low-rank properties, and applying the L2 norm for specific items to ensure connectivity: xT v =X v (C+D v )+E v in,‖ ‖ * represents the nuclear norm of the matrix, represents the square of the matrix bi-norm, ‖ ‖ 2,1 represents the 2,1 norm of the matrix, λ C ,λ D It is a parameter set to balance the regularization of the objective function; S25, finally get C and D v Then, construct the similarity matrix S: Perform spectral clustering on the matrix S to obtain the final clustering result.

2. According to claim 1, a 5G underground mine positioning method based on a convex relaxation constraint-based consistency-specific multi-view subspace clustering algorithm is characterized in that: The specific operation of combining the position information of each sampling point and the three signal strength parameter information into three d+2-dimensional high-dimensional data points in step S1 is: S11. First, define the location information of n sampling points as (x1, y1), (x2, y2), (x3, y3)..., (x n ,y n ), where n is the number of sampling points; S12. Each sampling point can obtain the signal strength of d 5GAPs, where the signal strength indication RSSI of the i-th sampling point is expressed as The signal received power RSRP is expressed as The signal reception quality RSRQ is expressed as S13. Finally, the position signal of each sampling point is combined with their three signal strength parameter information RSSI, RSRP, and RSRQ to obtain three d+2-dimensional high-dimensional data points, which are respectively expressed as:

3. According to claim 1, a 5G underground mine positioning method based on a convex relaxation constraint-based consistency-specific multi-view subspace clustering algorithm is characterized in that: The specific solution process of the consistent specific multi-view subspace clustering algorithm based on convex relaxation constraints in step S2 is: using convex relaxation and introducing intermediate variables to perform alternating optimization solutions, first introducing intermediate variables K and W v The formula in step S24 is converted into: xT v =X v (C+D v )+E v , C=K,E v =W v ; Then solve it by the method of augmented Lagrange multipliers: where <> is the standard Euclidean inner product of two matrices, ‖ ‖ F represents the norm of the matrix, Y1 v ,Y2,Y3 v is the Lagrange multiplier, and μ is the penalty term coefficient.

4. According to claim 1, a 5G underground mine positioning method based on a convex relaxation constraint-based consistency-specific multi-view subspace clustering algorithm is characterized in that: The specific implementation of step S3 is: S31. First, for each data point in each category of the collected samples, remove the position information (x i ,y i ) Only keep RSSI data i = {r i1 ,r i2 ,...,r id }; S32, then for each category, find the mean signal strength of the data points in these categories to obtain k mean points r k = {r k1 ,r k2 ,...,r kd }; S33, then calculate the Euclidean distance between the k mean points and the signal strength vector of the detected signal point, and obtain the category corresponding to the minimum value; S34. Finally, the position information of all vectors in this category is averaged to estimate the position of the detection point.

Images