A data processing method and device based on granular clustering

Through the granularity clustering method, the K-means and density peak clustering are used to quickly determine the K-means clustering and cluster center in big data scenarios, which solves the problems of low efficiency and poor accuracy of K-means clustering in traditional methods, and realizes the Fast and accurate clustering and real-time push of IPTV user behavior data.

CN113852845BActive Publication Date: 2024-11-26E-SURFING DIGITAL LIFE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110160712.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-02-05
Publication Date
2024-11-26
Estimated Expiration
2041-02-05

AI Technical Summary

Technical Problem

The existing data processing methods have a long processing time and low processing efficiency in big data scenarios. Especially in IPTV technology, the traditional K-means clustering method is sensitive to K values, has a great impact on initial value selection, and it is difficult to quickly determine the appropriate one. K value and initialize the cluster center, resulting in poor clustering effect and slow convergence speed.

Method used

Using a granularity clustering method, the original sample set is coarsely granulated through meshing, and the noise samples are eliminated using density peak clustering, the K value and estimated cluster center are quickly determined, and the cluster center is mapped to the fine-grained space for Clustering, and finally clustering is based on the K-means algorithm.

Benefits of technology

It significantly improves the speed and accuracy of clustering, reduces deviations during iteration, quickly converges, reduces the amount of calculation, and realizes real-time accurate processing of massive data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113852845B_ABST
    Figure CN113852845B_ABST
Patent Text Reader

Abstract

The present invention provides a data processing method and device based on granular clustering. The device includes: a data acquisition module, which is configured to collect user behavior data in real time as an original sample set; a granular processing module, which is configured to: remove noise samples based on coarse-grained division of the original sample set and density peak clustering, determine the number of original sample clusters and estimate the original sample cluster centers; and determine new sample sets and new cluster centers based on fine-grained division of the original sample set and original sample cluster center mapping; a data clustering module, which is configured to perform clustering based on the determined new sample sets and new cluster centers; and a data push module, which is configured to push relevant data to each user group associated with each cluster in real time based on the clustering results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big data, and more specifically, to a data processing method and device based on granular clustering. Background Art

[0002] With the rapid development of IPTV technology, the amount of information people receive when watching IPTV programs has increased dramatically, and higher requirements have been placed on the selection of program content. There are currently some data processing methods based on data mining and machine learning, but in the face of the ever-increasing amount of user behavior data, these methods have long processing time and low processing efficiency. How to achieve real-time and accurate push in big data scenarios has become an urgent problem to be solved.

[0003] In order to process massive data, clustering is required. In the traditional K-means clustering method, the K value has a crucial impact on the final result. Prior knowledge is required to give a suitable K value. Estimation out of thin air leads to poor clustering effect. K samples are randomly selected as the center of the cluster. The initial value of the center cluster is sensitive. Choosing different initial values ​​may lead to different cluster division rules and slow convergence. In addition, the K-means algorithm uses the mean of all points as the new center point during the iteration process. If there are abnormal points in the cluster, the mean deviation will be serious.

[0004] The Chinese patent application “A grid clustering algorithm based on density peak” (CN201710502536.7) proposes a grid clustering algorithm based on density peak, but the K value cannot be determined quickly and is sensitive to noise.

[0005] The Chinese patent application “A Grid-Based Density Peak Clustering Method and System” (CN201610515319.7) proposes a grid-based density peak clustering method, but it is impossible to quickly find a suitable K value and it is impossible to estimate the initialization cluster center.

[0006] Therefore, in order to quickly remove noise in big data scenarios, improve clustering accuracy and greatly increase clustering speed, it is hoped to provide an improved data processing method. Summary of the invention

[0007] This Summary is provided to introduce some concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0008] The present invention proposes a data processing method and device based on granular clustering. The method can quickly generate user grouping labels based on user behavior data, and push content of interest to different user groups in real time. Compared with traditional data processing methods, the processing speed and accuracy are greatly improved. More specifically, the method of the present invention first uses grid division to coarse-grain the original sample set, removes noise samples through density peak clustering, quickly determines the K value and estimates the original sample cluster center. Next, the original sample set is fine-grained, and the previously estimated cluster center is mapped to the fine-grained space. Finally, the new sample set transformed from the fine-grained is clustered.

[0009] According to one aspect of the present invention, a data processing method based on granular clustering is provided, the method comprising:

[0010] Collect user behavior data in real time as the original sample set;

[0011] Based on coarse-grained partitioning and density peak clustering of the original sample set, noise samples are eliminated, the number of original sample clusters is determined, and the centers of the original sample clusters are estimated;

[0012] Determine a new sample set and a new cluster center based on fine-grained division of the original sample set and mapping of the original sample cluster centers;

[0013] Perform clustering based on the determined new sample set and new cluster centers; and

[0014] Based on the clustering results, relevant data is pushed to each user group associated with each cluster in real time.

[0015] According to an embodiment of the present invention, performing coarse-grained division and density peak clustering on the original sample set further includes:

[0016] Determining a set of grid cells containing original samples in the coarse-grained grid space; and

[0017] Determine the density ρ of each grid cell in the set of grid cells i and the relative distance δ i To perform density peak clustering, where the density ρ i is the number of sample points in the grid unit, ρ i =|x 3i |, the relative distance δ i is the relative distance between the grid unit and other grid units, in d ij Represents the grid cell x 3i and x 3j Manhattan distance, d ij=|x 3i (1) -x 3j (1) |+|x 3i (2) -x 3j (2) |+…+|x 3i (T) -x 3j (T) |.

[0018] According to a further embodiment of the present invention, the density peak clustering further comprises:

[0019] The relative distance δ in the set of grid cells i Larger but density ρ i Smaller grid cells are considered as abnormal grid cells and removed;

[0020] Choose a suitable relative distance δ among the set of grid cells i Larger and with density ρ i The larger grid cell serves as the central grid cell; and

[0021] Determine the mean of the original sample points in each central grid unit to obtain the original sample cluster center.

[0022] According to a further embodiment of the present invention, the original sample cluster center mapping further includes:

[0023] The new sample set and the new cluster center are determined by mapping the original sample set after removing the noise samples and the estimated original sample cluster center into the fine-grained space.

[0024] According to a further embodiment of the present invention, clustering based on the determined new sample set and new cluster center further includes:

[0025] K-means clustering is performed based on each sample in the new sample set and the estimated K new cluster centers, so that the sum of squares of distances from each sample to its nearest cluster center is minimized.

[0026] According to another aspect of the present invention, there is provided a data processing device based on granular clustering, the device comprising:

[0027] A data collection module, wherein the data collection module is configured to collect user behavior data in real time as an original sample set;

[0028] A granularity processing module, wherein the granularity processing module is configured to:

[0029] Based on the coarse-grained division and density peak clustering of the original sample set to remove noise samples, the number of original sample clusters is determined and the centers of the original sample clusters are estimated; and

[0030] Determine a new sample set and a new cluster center based on fine-grained division of the original sample set and mapping of the original sample cluster centers;

[0031] A data clustering module, the data clustering module being configured to perform clustering based on the determined new sample set and new cluster centers; and

[0032] A data push module is configured to push relevant data to each user group associated with each cluster in real time based on the clustering result.

[0033] According to an embodiment of the present invention, performing coarse-grained division and density peak clustering on the original sample set further includes:

[0034] Determining a set of grid cells containing original samples in the coarse-grained grid space; and

[0035] Determine the density ρ of each grid cell in the set of grid cells i and the relative distance δ i To perform density peak clustering, where the density ρ i is the number of sample points in the grid unit, ρ i =|x 3i |, the relative distance δ i is the relative distance between the grid unit and other grid units, in d ij Represents the grid cell x 3i and x 3j Manhattan distance, d ij =|x 3i (1) -x 3j (1) |+|x 3i (2) -x 3j (2) |+…+|x 3i (T) -x 3j (T) |.

[0036] According to a further embodiment of the present invention, the density peak clustering further comprises:

[0037] The relative distance δ in the set of grid cells i Larger but density ρ iSmaller grid cells are considered as abnormal grid cells and removed;

[0038] Choose a suitable relative distance δ among the set of grid cells i Larger and with density ρ i The larger grid cell serves as the central grid cell; and

[0039] Determine the mean of the original sample points in each central grid unit to obtain the original sample cluster center.

[0040] According to a further embodiment of the present invention, the original sample cluster center mapping further includes:

[0041] The new sample set and the new cluster center are determined by mapping the original sample set after removing the noise samples and the estimated original sample cluster center into the fine-grained space.

[0042] According to a further embodiment of the present invention, clustering based on the determined new sample set and new cluster center further includes:

[0043] K-means clustering is performed based on each sample in the new sample set and the estimated K new cluster centers, so that the sum of squares of distances from each sample to its nearest cluster center is minimized.

[0044] Compared with the solutions in the prior art, the data processing method and device based on granular clustering provided by the present invention have at least the following advantages:

[0045] (1) Remove outliers and reduce the deviation of cluster center in the iterative process;

[0046] (2) Determine the appropriate K value based on the data distribution and estimate the center positions of the K clusters to achieve rapid convergence and reduce the number of iterations; and

[0047] (3) Refine the original samples by meshing and reduce the amount of computation in each iteration.

[0048] These and other features and advantages will become apparent by reading the following detailed description and by reference to the associated drawings.It is to be understood that the foregoing general description and the following detailed description are illustrative only and are not restrictive of the aspects of what is claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to understand the manner in which the above features of the present invention are used in detail, the above briefly summarized contents can be described in more detail with reference to various embodiments, some of which are shown in the accompanying drawings. However, it should be noted that the accompanying drawings only show some typical aspects of the present invention and should not be considered to limit its scope, because the description may allow for other equally effective aspects.

[0050] Figure 1 An exemplary architecture diagram of a data processing device based on granular clustering according to an embodiment of the present invention is shown.

[0051] Figure 2 A flow chart of a data processing method based on granular clustering according to an embodiment of the present invention is shown.

[0052] Figure 3 A flowchart of a grid-based granular K-means clustering algorithm according to an embodiment of the present invention is shown.

[0053] Figure 4 An example of a STING mesh structure according to an embodiment of the present invention is shown.

[0054] Figure 5a-5b An example of center point selection according to an embodiment of the present invention is shown.

[0055] Figure 6 An example of coarse-grained partitioning of a STING grid according to an embodiment of the present invention is shown.

[0056] Figure 7 An example of a decision diagram of a grid unit according to an embodiment of the present invention is shown.

[0057] Figure 8 An example of fine-grained partitioning of a STING grid according to an embodiment of the present invention is shown.

[0058] Fig. 9 An example of mapping the original sample cluster centers to a fine-grained grid space according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0059] The present invention will be described in detail below in conjunction with the accompanying drawings, and the features of the present invention will be further revealed in the following specific description.

[0060] Figure 1 FIG. 1 is an exemplary architecture diagram of a data processing device 100 based on granular clustering according to an embodiment of the present invention. Figure 1 As shown, the device 100 of the present invention includes: a data acquisition module 101 , a granularity processing module 102 , a data clustering module 103 and a data pushing module 104 .

[0061] The data collection module 101 can collect user behavior data in real time as an original sample set. As an example, the data collection module 101 can collect user behavior data of watching TV programs in real time as an original sample set, wherein all programs are first classified by type, such as TV news and information programs, TV talk shows, TV literary programs, TV entertainment programs, TV documentaries, TV dramas, TV movies, and TV special programs. At the same time, a TV program can have multiple types. The data collection module 101 counts the TV program history records watched by user i in the past 30 days every day, and for each program type, accumulates the corresponding viewing time, and normalizes the measurement into a score, namely, time. t / (time1+time2+…+time T ). To this end, regarding N users, a matrix as shown in the following table is constructed, which is the original sample set D1.

[0062] Type 1 Type 2 Type 3 Type 4 … Type T User 1 0.15 0.01 0.23 0.00 … 0.10 User 2 0.05 0.30 0.01 0.00 … 0.20 User 3 0.10 0.25 0.17 0.30 … 0.00 … … … … … … … User N 0.35 0.05 0.16 0.02 … 0.10

[0063] The granularity processing module 102 can perform coarse-grained division on the original sample set D1 obtained from the data acquisition module 101, remove noise samples through density peak clustering, quickly determine the number of original sample clusters (for example, K value) and estimate the original sample cluster centers. The granularity processing module 102 can also perform fine-grained division on the original sample set D1, and map the previously estimated cluster centers to the fine-grained space to form a new sample format. The coarse-grained division and the fine-grained division can be performed using the STING grid (statistical information grid). Figure 4 An example of a STING grid structure 400 according to an embodiment of the present invention is shown. STING is a grid-based multi-resolution clustering technology that divides a spatial region into rectangular cells. For different levels of resolution, there are usually multiple levels of rectangular cells, which form a hierarchical structure, in which each cell at a higher level is divided into multiple cells at a lower level.

[0064] The data clustering module 103 may perform clustering based on the fine-grained new sample set and the new cluster centers. In some cases, clustering may be performed using, for example, a K-means clustering algorithm.

[0065] The data push module 104 can push relevant data to each user group in real time according to the clustering results. In one example, the TV users can be automatically divided into K groups through the clustering algorithm, and then the T attributes (program types) of the cluster center of each group are sorted, and the background pushes relevant programs to each group according to their respective Top-N attributes (program types).

[0066] For the convenience of explanation, the embodiments of the present invention will be described below by taking the granular K-means clustering algorithm based on the STING grid as an example, but those skilled in the art will appreciate that the present invention is also applicable to other clustering algorithms.

[0067] Figure 2 2 is a flow chart of a data processing method 200 based on granular clustering according to an embodiment of the present invention. The method starts at step 201, where the data collection module 101 collects user behavior data in real time as an original sample set D1.

[0068] In step 202, the granularity processing module 102 performs coarse-grained division on the collected original sample set D1, removes noise samples, determines the number of original sample clusters and estimates the original sample cluster centers, wherein the number of original sample clusters (in the case of the K-means clustering algorithm, the K value) and the original sample cluster centers can be quickly determined by density peak clustering. Specifically, the granularity processing module 102 first uses STING grid division to coarse-grain the original sample set D1, with a scale parameter of ε1, to obtain the grid space GS1. In layman's terms, assuming that the original sample x1 has T attributes, that is, {a 11 ,a 12 ,…,a 1T}, then for each a 1t The dimensions are normalized and divided into ε1 equal parts Then, the set of grid cells containing original samples in the current grid space GS1 is defined as sample set D2, the entire sample set D2 is scanned, and the number of original sample points in the grid cell is used as the density of the grid cell. The relative distance of each grid cell is calculated by using the idea that the distance between the cluster center grid cell and other cluster center grid cells is large, and the distance between the grid cell and other grid cells in its grid cell cluster is small. The algorithm steps are as follows:

[0069] (1) Calculate the density ρ of the grid cells i , that is, the number of data points in grid cell i, where:

[0070] v i =|x 3i |

[0071] (2) Calculate the distance δ of the grid cells i . Sort the grid cells in descending order of frequency, where δ i is the relative distance of sample point i:

[0072]

[0073] Among them, the indicator set Obviously when Sometimes, there is dij represents the sample point x in grid cells i and j 3i and x 3j Manhattan distance, d ij It is calculated by the following formula:

[0074] d ij =|x 3i (1) -x 3j (1) |+|x 3i (2) -x 3j (2) |+…+|x 3i (T) -x 3j (T) |

[0075] Then, according to the calculated density ρ of the grid cells i and distance δ i To obtain the decision graph. Specifically, the relative distance δ i Larger but density ρ i Smaller grid cells are regarded as abnormal grid cells and removed, and the appropriate relative distance δ is selected. i Larger and with density ρ i The larger grid unit is used as the central grid unit, denoted as c 2k . Query the mean of the original sample points in each central grid unit respectively, as the cluster center c of the original sample 1k Regarding the coarse-grained division, the number of samples is greatly reduced compared to the original one. Secondly, no matter how large the original data is, the number of grid cells remains unchanged. Figure 5a-5b It can be seen that based on the scatter plot of the original sample x1, the decision graph is constructed by the traditional density peak clustering algorithm to obtain the K value 2 and the cluster center x 11 、x 110 , and the noise sample x to be removed 128 . Compared with the process of constructing decision graphs using traditional density peak clustering algorithms (such as Figure 5a-5b ), especially in big data scenarios, this algorithm can significantly reduce the amount of calculation. This approach does not obtain the cluster center as accurately as the traditional density peak clustering algorithm, but it can quickly determine the K value and estimate the approximate location of the cluster center.

[0076] In step 203, the granularity processing module 102 performs a fine-grained partitioning on the original sample set D1, and maps the original sample set D1 after removing noise samples and the estimated original sample cluster centers into the fine-grained space, forming a new sample set D3 and new cluster centers. Specifically, the granularity processing module 102 can use STING grid partitioning to perform fine-graining on the original sample set D1. The partitioning process is the same as that of coarse-graining, but the scale parameter becomes ε2 (ε1 << ε2), and finally the grid space GS2 is obtained. The set of grid cells corresponding to each original sample mapped into the grid space GS2 is called the sample set D3. In the sample set D3, the original samples belonging to the same grid cell are regarded as new samples x3 with the same attributes. In addition, the number of different samples x3 in the sample set D3 is denoted as M. Since the grid space is fixed, its number is denoted as L and M <= L. As the scale N of the original data increases, M is at most L, but L << N. That is to say, no matter how much the scale of the original sample set D1 grows later, the O(N - M) repeated calculations for finding the distance between the sample x3 and the cluster center during each iteration of the sample set D3 to be clustered can be reduced. Subsequently, K cluster center points c of the original samples can be found 1k of the corresponding grid cells in the grid space GS2 as the new K cluster centers c 3k .

[0077] In step 204, the data clustering module 103 performs clustering based on the fine-grained new sample set and the new cluster centers. Specifically, the data clustering module 103 can perform K-means clustering according to each sample x in the new sample set D3 3i and the estimated K cluster centers c 3k to minimize the sum of the squares of the distances from each sample point to its nearest cluster center, and use such sum of the squares of the distances as the objective function, denoted as J. The expression of this objective function is as follows:

[0078]

[0079] In Figure 3 an example of the grid-based granularity clustering algorithm 300 is further described in detail. As Figure 3 shown, after obtaining the original sample set D1 including N original samples x1, noise data is removed and the value of K and the cluster centers c are estimated based on the coarsened grid and density peak clustering method 1k . Subsequently, based on the fine-grained grid, each original sample x1 after removing noise is re-converted into a new sample x3, and the K cluster centers c of the original samples 1k are converted into new cluster centers c 3kThen, the K-means clustering algorithm is used to divide the new sample x3 into the nearest cluster, and the center of each cluster is recalculated. The process is repeated until convergence to obtain the clustering results of each cluster.

[0080] return Figure 2 After obtaining the clustering results, in step 205, the data push module 104 can push relevant data to each user group associated with each cluster in real time according to the clustering results.

[0081] Figure 6-9 The implementation of the key parts of the granular K-means algorithm according to one embodiment is further described. Figure 5a The scatter plot of the original sample x1 shown in is first coarse-grained, as Figure 6 As shown, the scale parameter ε1 is 4, and the STING grid is divided into 16 grid cells (i.e., samples x2). For example, x 21 is the grid cell with coordinates (1,1), x 22 is the grid cell with coordinates (1,2), and so on. 216 is the grid unit with coordinates (4,4). Through the coarse-grained grid units, according to the improved set rules, the decision diagram is constructed as follows Figure 7 As shown, based on this figure, the noise grid unit x is removed 215 , determine the K value to be 2 and x 26 、x 28 The center grid unit is the center grid unit, and the mean of the original sample points in each center grid unit is queried respectively as the center of the original sample cluster (solid gray), as shown in Figure 6 As shown. Since the grid unit x 23 , grid cell x 211 , grid cell x 212 , grid cell x 213 , grid cell x 214 , grid cell x 216 There is no original sample, and there is no need to calculate the corresponding ρ and δ. The final number of x2 samples that need to be processed is 10.

[0082] After the coarse-graining process, the original sample set is fine-grained, such as Figure 8 As shown in, the original sample set after removing the noise samples is mapped to the fine-grained grid space with a scale parameter ε2 of 16, and then as Fig. 9 As shown in FIG. 2 , the center points of the two original sample clusters previously determined are mapped to the fine-grained grid space. Fig. 9 The final new sample set D3 to be clustered and the initialized cluster center c are shown in 31 、c 32 .

[0083] What has been described above includes examples of various aspects of the claimed subject matter. Of course, it is not possible to describe every conceivable combination of components or methods for the purpose of describing the claimed subject matter, but one of ordinary skill in the art will recognize that many further combinations and permutations of the claimed subject matter are possible. Thus, the disclosed subject matter is intended to encompass all such alterations, modifications, and variations that fall within the spirit and scope of the appended claims.

Claims

1. A data processing method based on granular clustering, It is characterized in that The method comprises: Collect user behavior data in real time as the original sample set; Based on coarse-grained partitioning and density peak clustering of the original sample set, noise samples are eliminated, the number of original sample clusters is determined, and the centers of the original sample clusters are estimated; Determine a new sample set and a new cluster center based on fine-grained division of the original sample set and mapping of the original sample cluster centers; Perform clustering based on the determined new sample set and new cluster centers; and Based on the clustering results, relevant data is pushed to each user group associated with each cluster in real time.

2. The method according to claim 1, It is characterized in that The coarse-grained division and density peak clustering of the original sample set further includes: Determining a set of grid cells containing original samples in the coarse-grained grid space; and Determine the density ρ of each grid cell in the set of grid cells i and the relative distance δ i To perform density peak clustering, where the density ρ i is the number of sample points in the grid unit, ρ i =|x 3i |, the relative distance δ i is the relative distance between the grid unit and other grid units, in d ij represents the sample point x in grid cells i and j 3i and x 3j Manhattan distance, d ij =|x 3i (1) -x 3j (1) |+|x 3i (2) -x 3j (2) |+…+|x 3i (T) -x 3j (T) |, where T represents the sample point x 3i and x 3j The number of attributes.

3. The method according to claim 2, It is characterized in that The density peak clustering further comprises: The relative distance δ in the set of grid cells i Greater than the predetermined distance threshold but density ρ i Grid cells with density less than a predetermined threshold are considered as abnormal grid cells and removed; Choose the relative distance δ among the set of grid cells i is greater than the predetermined distance threshold and the density ρ i The grid cells with a density greater than the predetermined density threshold are used as central grid cells; and Determine the mean of the original sample points in each central grid unit to obtain the original sample cluster center.

4. The method according to claim 1, It is characterized in that The original sample cluster center mapping further includes: The new sample set and the new cluster center are determined by mapping the original sample set after removing the noise samples and the estimated original sample cluster center into the fine-grained space.

5. The method according to claim 1, It is characterized in that Clustering based on the determined new sample set and new cluster center further includes: K-means clustering is performed based on each sample in the new sample set and the estimated K new cluster centers, so that the sum of squares of distances from each sample to its nearest cluster center is minimized.

6. A data processing device based on granular clustering, It is characterized in that The device comprises: A data collection module, wherein the data collection module is configured to collect user behavior data in real time as an original sample set; A granularity processing module, wherein the granularity processing module is configured to: Based on the coarse-grained division and density peak clustering of the original sample set to remove noise samples, the number of original sample clusters is determined and the centers of the original sample clusters are estimated; and Determine a new sample set and a new cluster center based on fine-grained division of the original sample set and mapping of the original sample cluster centers; A data clustering module, the data clustering module being configured to perform clustering based on the determined new sample set and new cluster centers; and A data push module is configured to push relevant data to each user group associated with each cluster in real time based on the clustering result.

7. The device according to claim 6, It is characterized in that The coarse-grained division and density peak clustering of the original sample set further includes: Determining a set of grid cells containing original samples in the coarse-grained grid space; and Determine the density ρ of each grid cell in the set of grid cells i and the relative distance δ i To perform density peak clustering, where the density ρ i is the number of sample points in the grid unit, ρ i =|x 3i |, the relative distance δ i is the relative distance between the grid unit and other grid units, in d ii represents the sample point x in grid cells i and j 3i and x 3j Manhattan distance, d ij =|x 3i (1) -x 3j (1) |+|x 3i (2) -x 3j (2) |+…+|x 3i (T) -x 3j (T) |, where T represents the sample point x 3i and x 3j The number of attributes.

8. The device according to claim 7, It is characterized in that The density peak clustering further comprises: The relative distance δ in the set of grid cells i Greater than the predetermined distance threshold but density ρ i Grid cells with density less than a predetermined threshold are considered as abnormal grid cells and removed; Choose the relative distance δ among the set of grid cells i Greater than a predetermined distance threshold and density ρ i Grid cells with a density greater than a predetermined threshold are regarded as central grid cells; and Determine the mean of the original sample points in each central grid unit to obtain the original sample cluster center.

9. The device according to claim 6, It is characterized in that The original sample cluster center mapping further includes: The new sample set and the new cluster center are determined by mapping the original sample set after removing the noise samples and the estimated original sample cluster center into the fine-grained space.

10. The device according to claim 6, It is characterized in that Clustering based on the determined new sample set and new cluster center further includes: K-means clustering is performed based on each sample in the new sample set and the estimated K new cluster centers, so that the sum of squares of distances from each sample to its nearest cluster center is minimized.