A method for electromagnetic spectrum situation reinforcement based on K-means clustering algorithm

Abstract

The present invention proposes an electromagnetic spectrum situation reinforcement method based on the K-means clustering algorithm, which relates to the field of electromagnetic spectrum management and situation awareness, and aims to effectively complete electromagnetic spectrum data through unsupervised learning, and overcome the dependence on a large number of samples and the problems of sample imbalance and migration in traditional methods. The present invention first divides the target area into grids, and collects electromagnetic spectrum data of each grid point through a measuring device. Then, the elbow method is used to determine the optimal number of clusters for K-means clustering, and the collected data is clustered and analyzed to obtain the cluster center of the spectrum situation. Then, based on the clustering results, the missing spectrum data is completed to generate a complete spectrum situation map. The accuracy of the completed data is verified by similarity evaluation. The present invention reduces the computational complexity, improves the accuracy and reliability of spectrum situation awareness, and is suitable for real-time management and optimization of spectrum resources.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic spectrum management and situation awareness, and in particular to an electromagnetic spectrum situation reinforcement method based on a K-means clustering algorithm, which aims to complete electromagnetic spectrum data and reconstruct the situation through an efficient unsupervised learning algorithm. Background Art

[0002] With the increasing complexity of the electromagnetic environment and the increasing shortage of spectrum resources, how to effectively perceive and manage the electromagnetic spectrum situation has become an important research topic. At present, the commonly used spectrum situation awareness methods mainly rely on deep neural networks and graph neural networks, which often require a large number of labeled samples for supervised learning. In the "Electromagnetic Spectrum Map Construction Method Based on Residual Encoder" published in the journal "Radio Communication Technology" by Zhang Han and others from the School of Information and Communication Engineering of Harbin Engineering University, the residual encoder is trained by using open source spectrum historical data sets, combined with some observation data of the target area, to achieve the completion and reconstruction of electromagnetic spectrum data. In the "Spectrum Prediction Algorithm Based on GCN-LSTM" published in the journal "Radio Communication Technology" by Xue Wenju and others from the Institute of Communication Technology of Harbin Institute of Technology, based on graph convolutional neural networks, combined with long short-term memory networks, and introducing self-attention mechanisms, the electromagnetic spectrum situation is predicted based on a large amount of past data and some observation data. In the national invention patent "An Intelligent Electromagnetic Spectrum Map Fusion Method Based on Dense Autoencoder" applied by Chang Xin and others from the 54th Institute of China Electronics Technology Group Corporation, with the authorization announcement number CN118070230B, the dense autoencoder is trained using relevant historical data, and then the electromagnetic spectrum map is fused in combination with some observation data. However, in practical applications, the number of samples is limited and unevenly distributed, which leads to the problem of sample imbalance in the training process of traditional methods and difficulty in migration in different environments. In the prior art, although deep learning based on supervised learning has achieved certain success in dealing with spectrum situation awareness problems under some conditions, they are highly dependent on the quality and quantity of samples, which increases the complexity and computational cost of the model. In addition, in terms of transfer learning, traditional methods also have the problems of insufficient adaptability and poor generalization ability. Therefore, there is an urgent need for a new method that does not rely on a large number of samples and can effectively handle spectrum situation awareness and completion in different environments. Summary of the invention

[0003] In view of the deficiencies of the prior art, the present invention proposes an electromagnetic spectrum situation reinforcement method based on the K-means clustering algorithm, which avoids the dependence on a large number of samples and does not require complex training using graph neural networks or adversarial neural networks, thus overcoming the problems of sample imbalance and migration. The present invention performs situation completion on electromagnetic spectrum data through the unsupervised K-means clustering algorithm, effectively improving the accuracy and reliability of spectrum situation awareness.

[0004] The technical solution adopted by the present invention is as follows:

[0005] An electromagnetic spectrum situation reinforcement method based on the K-means clustering algorithm, comprising the following steps:

[0006] Step 1: Perform grid processing on the target area, and use multiple electromagnetic spectrum sensing devices to measure the electromagnetic spectrum intensity of the grids where they are located respectively, forming an electromagnetic spectrum situation awareness matrix ; where the number of electromagnetic spectrum sensing devices is less than the number of grids;

[0007] Step 2: Use the K-means algorithm to cluster the electromagnetic spectrum situation awareness matrix to obtain a set of clustering centers; the set of clustering centers contains multiple clustering center positions and the corresponding electromagnetic spectrum intensities;

[0008] Step 3: Obtain the grid positions not measured by the electromagnetic spectrum sensing devices, traverse the set of clustering centers, find the clustering center with the smallest distance from the clustering center position, use the electromagnetic spectrum intensity corresponding to the clustering center with the smallest distance as the reinforcement data, and superimpose the grid positions to form an electromagnetic spectrum situation estimation matrix , and the electromagnetic spectrum situation estimation matrix is merged with the electromagnetic spectrum situation awareness matrix to obtain an electromagnetic spectrum situation enhancement matrix .

[0009] Further, the specific process of Step 1 is as follows:

[0010] Perform grid processing on the target area using a square, set the side length of each grid to , set the number of rows and columns of the network to N and M respectively; use P electromagnetic spectrum sensing devices to measure the electromagnetic spectrum intensity of the grids where they are located respectively, where P < NM, and the positions corresponding to the electromagnetic spectrum sensing devices in the grids are and , 1 ≤ n ≤ N, 1 ≤ m ≤ M; divide the target frequency range into L continuous frequency bands, record the electromagnetic spectrum intensity measured by each frequency band and grid as , and fill the unmeasured grids with zeros to form an electromagnetic spectrum situation awareness matrix :

[0011] D = [ [ x 1 , y 1 , a 1 , 1 , 1 , ⋯ , a l , 1 , 1 , ⋯ , a L , 1 , 1 ] ⋮ [ x n , y m , a 1 , n , m , ⋯ , a l , n , m , ⋯ , a L , n , m ] ⋮ [ x N , y M , a 1 , N , M , ⋯ , a l , N , M , ⋯ , a L , N , M ] ] P × ( L + 2 ) .

[0012] Furthermore, the specific process of step 2 is:

[0013] Step 2-1: Electromagnetic Spectrum Situational Awareness Matrix Use K-means algorithm for clustering, adopt elbow method, traverse the sum of squared errors within clusters in the range of [1, K], and take the number of clusters as the variable, and take the number of clusters corresponding to the maximum value of the second derivative of the sum of squared errors within clusters as the result; K is the set value;

[0014] Step 2-2: Repeat step 2-1 E times to construct the cluster number result K = [ K 1 , ⋯ , K e , ⋯ , K E ] , and then find the optimal mean ; Where, E is the set value;

[0015] ;

[0016] Step 2-3: Set the clustering value of the K-means method to , the electromagnetic spectrum situational awareness matrix Perform clustering and obtain the cluster center set , which contains The location of cluster centers and the corresponding electromagnetic spectrum intensity c = [ x ′ c , y ′ c , a ′ 1 , c , ⋯ , a ′ l , c , ⋯ , a ′ L , c ] 1 × ( L + 2 ) ; is a cluster center location, is the electromagnetic spectrum intensity of the l-band corresponding to the cluster center.

[0017] Furthermore, the specific process of step 3 is as follows:

[0018] Step 3-1: Obtain grid locations not measured by electromagnetic spectrum sensing equipment and ;

[0019] Step 3-2: Traverse the cluster center set ,turn up The smallest cluster center generates reinforcement data [ x n , y m , a ′ 1 , c , ⋯ , a ′ l , c , ⋯ , a ′ L , c ] 1 × ( L + 2 ) , put into the electromagnetic spectrum situation assessment matrix middle;

[0020] Step 3-3: Repeat steps 3-1 and 3-2 until all grids not measured by the electromagnetic spectrum sensing device are generated with reinforcement data, and the electromagnetic spectrum situation estimation matrix and the Electromagnetic Spectrum Situational Awareness Matrix Merge to get the electromagnetic spectrum situation enhancement matrix .

[0021] The advantages of the present invention compared with the prior art are:

[0022] By introducing the K-means clustering algorithm, the present invention effectively avoids the dependence of graph neural networks and adversarial neural networks on a large number of samples when dealing with electromagnetic spectrum situational awareness problems, avoids sample imbalance and transfer learning problems, and reduces computational complexity and resource consumption. The specific beneficial effects are as follows:

[0023] 1. Reduce dependence on sample quantity and quality: Using unsupervised learning methods, there is no need to rely on a large number of labeled samples, which reduces the difficulty and cost of data acquisition.

[0024] 2. Improve generalization ability and adaptability: Without the need for complex transfer learning process, spectrum situation reinforcement can be performed directly in different environments, which enhances the generalization ability of the method.

[0025] 3. Improve computational efficiency: The K-means clustering algorithm is used, which has low computational complexity and is suitable for real-time spectrum situation awareness and reinforcement.

[0026] 4. Enhance the accuracy of spectrum situation awareness: By completing missing data, the completeness and accuracy of the spectrum situation map are improved, which helps to effectively manage spectrum resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of an electromagnetic spectrum situation reinforcement method based on a K-means clustering algorithm according to an embodiment of the present invention.

[0028] Figure 2 This is the electromagnetic spectrum perception situation collected from frequency band 1.

[0029] Figure 3 This is the electromagnetic spectrum perception situation collected from frequency band 2.

[0030] Figure 4 This is the electromagnetic spectrum perception situation collected from frequency band 3.

[0031] Figure 5 This is the electromagnetic spectrum perception situation collected from frequency band 4.

[0032] Figure 6 This is the electromagnetic spectrum perception situation collected from frequency band 5.

[0033] Figure 7 This is the complete situation of the electromagnetic spectrum collected from frequency band 1.

[0034] Figure 8 This is the complete situation of the electromagnetic spectrum collected from frequency band 2.

[0035] Fig. 9 The complete electromagnetic spectrum situation collected from frequency band 3.

[0036] Fig.10 The complete electromagnetic spectrum situation collected from frequency band 4.

[0037] Fig.11 The complete electromagnetic spectrum situation collected from frequency band 5.

[0038] Fig.12 Schematic diagram of the change of the sum of squared errors within the cluster of invention embodiments with the number of clusters.

[0039] Fig.13 Schematic diagram of the change of the number of clusters of invention embodiments with the number of optimization times.

[0040] Fig.14 The enhanced situation of the electromagnetic spectrum collected from frequency band 1.

[0041] Fig.15 The enhanced situation of the electromagnetic spectrum collected from frequency band 2.

[0042] Fig.16 The enhanced situation of the electromagnetic spectrum collected from frequency band 3.

[0043] Fig.17 The enhanced situation of the electromagnetic spectrum collected from frequency band 4.

[0044] Fig.18 The enhanced situation of the electromagnetic spectrum collected from frequency band 5. Detailed implementation manners

[0045] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0046] An electromagnetic spectrum situation enhancement method based on the K-means clustering algorithm, the overall process is as Figure 1 shown, and specifically includes the following steps:

[0047] Step 1: Electromagnetic situation data collection;

[0048] The target area is processed into a grid using a square. For each grid, the side length is set to = 1 km, and the number of rows and columns of the network are set to N = 10 and M = 10 respectively. P = 70 electromagnetic spectrum sensing devices are used for measurement, where P < NM. The positions corresponding to the electromagnetic spectrum sensing devices in the grid are and , 1 ≤ n ≤ N, 1 ≤ m ≤ M; the target frequency range is divided into L = 5 continuous frequency bands, and the electromagnetic spectrum intensity measured for each frequency band and grid is recorded as , and fill the unmeasured grids with zeros to form an electromagnetic spectrum situational awareness matrix ; The electromagnetic spectrum perception situation collected is divided into frequency bands as follows Figures 2 to 6 As shown;

[0049] D = [ [ x 1 , y 1 , a 1 , 1 , 1 , ⋯ , a l , 1 , 1 , ⋯ , a L , 1 , 1 ] ⋮ [ x n , y m , a 1 , n , m , ⋯ , a l , n , m , ⋯ , a L , n , m ] ⋮ [ x N , y M , a 1 , N , M , ⋯ , a l , N , M , ⋯ , a L , N , M ] ] P × ( L + 2 ) .

[0050] In order to verify the effectiveness of the algorithm, all 100 cells were measured to obtain complete electromagnetic spectrum situation information, that is, the complete electromagnetic spectrum situation matrix The complete electromagnetic spectrum is divided into frequency bands as follows: Figures 7 to 11 shown.

[0051] Step 2: Clustering of electromagnetic situation data;

[0052] Step 2-1: Obtaining the number of clusters based on the K-means method;

[0053] Electromagnetic Spectrum Situational Awareness Matrix The K-means algorithm is used for clustering, and the elbow method is adopted to traverse the intra-cluster error sum of squares in the range of [1, K], where K is 20; the number of clusters is used as a variable, and the cluster number corresponding to the maximum value of the second derivative of the intra-cluster error sum of squares is taken as the result. Fig.12 Schematic diagram of the change of the sum of squared errors within clusters with the number of clusters.

[0054] Step 2-2: Calculate the mean of cluster number;

[0055] Repeat step 2-1 for a total of E times, where E is 1000, to construct the cluster number result K = [ K 1 , ⋯ , K e , ⋯ , K E ] , and then find the optimal mean =9. Fig.13 It is a schematic diagram of the change of cluster number with the number of optimization times;

[0056] .

[0057] Step 2-3: Get the cluster center;

[0058] Set the clustering value of the K-means method to , the electromagnetic spectrum situational awareness matrix Perform clustering and obtain the cluster center set , which contains The location of cluster centers and the corresponding electromagnetic spectrum intensity c = [ x ′ c , y ′ c , a ′ 1 , c , ⋯ , a ′ l , c , ⋯ , a ′ L , c ] 1 × ( L + 2 ) ; is a cluster center location, is the electromagnetic spectrum intensity of the l-band corresponding to the cluster center.

[0059] Step 3: Enhance electromagnetic situational data;

[0060] Step 3-1: Obtaining unmeasured grid positions;

[0061] Get grid locations that are not measured by electromagnetic spectrum sensing equipment and .

[0062] Step 3-2: Generate reinforcement data;

[0063] Traversing cluster centers ,turn up The smallest cluster center generates reinforcement data [ x n , y m , a ′ 1 , c , ⋯ , a ′ l , c , ⋯ , a ′ L , c ] 1 × ( L + 2 ) , put into the electromagnetic spectrum situation assessment matrix middle.

[0064] Step 3-3: Repeat steps 3-1 and 3-2 until all grids not measured by the electromagnetic spectrum sensing device are generated with reinforcement data, and the electromagnetic spectrum situation estimation matrix and the Electromagnetic Spectrum Situational Awareness Matrix Merge to get the electromagnetic spectrum situation reinforcement matrix . Figures 14 to 18 It is the reinforcement situation of the electromagnetic spectrum collected according to the frequency band division.

[0065] Furthermore, the electromagnetic spectrum situation reinforcement matrix Complete electromagnetic spectrum situation matrix The obtained cosine similarity is 99.96%, indicating that this method can accurately reinforce the electromagnetic spectrum situation without the need for historical data training.

Claims

1. A method for reinforcing electromagnetic spectrum situation based on K-means clustering algorithm, characterized in that: The following steps are involved: Step 1: Grid the target area and use multiple electromagnetic spectrum sensing devices to measure the electromagnetic spectrum intensity of the grid to form an electromagnetic spectrum situational awareness matrix. ; The number of electromagnetic spectrum sensing devices is less than the number of grids; Step 2: Electromagnetic Spectrum Situational Awareness Matrix Use K-means algorithm to cluster and get the cluster center set ; The cluster center set includes multiple cluster center locations and corresponding electromagnetic spectrum intensities; Step 3: Obtain the grid locations that are not measured by the electromagnetic spectrum sensing device and traverse the cluster center set , find the cluster center with the smallest distance from the cluster center, use the electromagnetic spectrum intensity corresponding to the smallest distance cluster center as reinforcement data, and superimpose the grid positions to form the electromagnetic spectrum situation estimation matrix , the electromagnetic spectrum situation assessment matrix Electromagnetic Spectrum Situational Awareness Matrix Merge to get the electromagnetic spectrum situation enhancement matrix ; Among them, the specific process of step 1 is: The target area is meshed with squares, and the side length of each grid is set to , the number of rows and columns of the grid are set to N and M respectively; P electromagnetic spectrum sensing devices are used to measure the electromagnetic spectrum intensity of the grid where they are located, where P < NM, and the corresponding positions of the electromagnetic spectrum sensing devices in the grid are and , 1 ≤ n ≤ N, 1 ≤ m ≤ M; the target frequency range is divided into L continuous frequency bands, and the electromagnetic spectrum intensity measured by each frequency band and grid is recorded as , and the grids that are not measured are filled with zeros to form an electromagnetic spectrum situation awareness matrix : ； Among them, the specific process of step 3 is: Step 3-1: Obtain grid locations not measured by the electromagnetic spectrum sensing device and ; Step 3-2: Traverse the cluster center set ,turn up The smallest cluster center generates reinforcement data , put into the electromagnetic spectrum situation assessment matrix Among them, is a cluster center location, is the electromagnetic spectrum intensity of the l-band corresponding to the cluster center; Step 3-3: Repeat steps 3-1 and 3-2 until all grids not measured by the electromagnetic spectrum sensing device are generated with reinforcement data, and the electromagnetic spectrum situation estimation matrix and the Electromagnetic Spectrum Situational Awareness Matrix Merge to get the electromagnetic spectrum situation enhancement matrix .

2. According to the method for reinforcing electromagnetic spectrum situation based on K-means clustering algorithm according to claim 1, it is characterized in that: The specific process of step 2 is as follows: Step 2-1: Electromagnetic Spectrum Situational Awareness Matrix Use K-means algorithm for clustering, adopt elbow method, traverse the sum of squared errors within clusters in the range of [1, K], and take the number of clusters as the variable, and take the number of clusters corresponding to the maximum value of the second derivative of the sum of squared errors within clusters as the result; K is the set value; Step 2-2: Repeat step 2-1 E times to construct the cluster number result , and then find the optimal mean ; Where, E is the set value; ； Step 2-3: Set the clustering value of the K-means method to , the electromagnetic spectrum situational awareness matrix Perform clustering and obtain the cluster center set , which contains The location of cluster centers and the corresponding electromagnetic spectrum intensity ; is a cluster center location, is the electromagnetic spectrum intensity of the l-band corresponding to the cluster center.

Images