A pulse clustering method and device based on WKMeans

Abstract

The present application discloses a pulse clustering method and device based on WKMeans, which first pre-processes the original pulse to obtain a pre-processed pulse, initializes the DOA parameters of the pre-processed pulse in spatial characteristics and directional characteristics to obtain the estimated K value of the clustering tree diagram, and then calculates the probability density of each data point in the mixed Gaussian distribution GMM to obtain the spatial distribution weight, adds it to the distance metric of the WKmeans algorithm to perform KMeans clustering to obtain a weighted clustering result, iterates and updates the cluster center, the category to which the sample point belongs, and the feature weight until the weighted clustering result converges or reaches a preset number of iterations, evaluates the weighted clustering result to obtain an evaluation result, and repeats the above steps until the preset requirements are met if the preset requirements are not met. By introducing DOA unsupervised hierarchical clustering, PDW weights and parameter normalization spatial distance optimization strategies in the iterative process of KMeans, accurate clustering division of electromagnetic radiation pulse data is achieved.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a WKMeans-based pulse clustering method and device. Background Art

[0002] At present, the K-means algorithm, as a classic unsupervised learning clustering method, plays an important role in many information clustering tasks. However, with the rapid increase in the scale of electromagnetic radiation source data and the increase in scene complexity, the K-means algorithm has exposed problems such as sensitivity to the initial cluster center and easy to fall into the local optimal solution when dealing with such problems. This type of traditional K-means clustering algorithm will also encounter performance bottlenecks when dealing with large-scale and complex scene data sets due to the sensitivity of the initial cluster center selection algorithm to outliers.

[0003] In the prior art, the K-means clustering algorithm completes clustering by iteratively searching for k center points in a data set and assigning data points to the cluster where the nearest center point is located. When processing data containing multiple features, all features are usually given the same weight. This results in that in some practical applications, features of different importance are treated equally in the clustering process, affecting the accuracy and effectiveness of the clustering effect. Summary of the invention

[0004] The purpose of this application is to overcome the existing technical defects and provide a pulse clustering method and device based on WKMeans, which realizes the accurate clustering division of electromagnetic radiation pulse data by introducing DOA unsupervised hierarchical clustering, PDW weight and parameter normalization spatial distance optimization strategy in the iterative process of KMeans.

[0005] The purpose of this application is achieved through the following technical solutions:

[0006] In a first aspect, the present application proposes a pulse clustering method based on WKMeans, the method comprising:

[0007] Step S1, preprocessing the original pulse to obtain a preprocessed pulse;

[0008] Step S2, performing initial hierarchical clustering on the DOA parameters of the preprocessed pulse in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram;

[0009] Step S3, based on the estimated K value, calculate the probability density of each data point in the mixed Gaussian distribution GMM to obtain the spatial distribution weight;

[0010] Step S4, adding the spatial distribution weight to the distance metric of the WKmeans algorithm and performing KMeans clustering to obtain a weighted clustering result;

[0011] Step S5, iteratively updating the cluster center, the category to which the sample point belongs, and the feature weight based on the weighted clustering result until the weighted clustering result converges or reaches a preset number of iterations;

[0012] Step S6: Evaluate the weighted clustering result to obtain an evaluation result. If the evaluation result does not meet the preset requirements, repeat steps S2 to S5 until the preset requirements are met.

[0013] In a possible implementation, step S4, adding spatial distribution weights to the distance metric of the WKmeans algorithm and performing KMeans clustering to obtain a weighted clustering result, includes:

[0014] Step S4-1, adding the spatial distribution weight to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance;

[0015] Step S4-2: Use the weighted Euclidean distance to perform KMeans clustering to obtain a weighted clustering result.

[0016] In a possible implementation, the preprocessing includes data cleaning to eliminate noise effects, feature transformation, normalization, and standardization.

[0017] In a possible implementation, the spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

[0018] In a possible implementation, the estimated K value is used to provide initial cluster number and spatial distribution information.

[0019] In a second aspect, the present application proposes a WKMeans-based pulse clustering device, the device comprising:

[0020] A preprocessing module, used for preprocessing the original pulse to obtain a preprocessed pulse;

[0021] The first clustering module is used to initialize hierarchical clustering of the DOA parameters of the pre-processed pulses in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram;

[0022] A calculation module is used to calculate the probability density of each data point in the mixed Gaussian distribution GMM based on the estimated K value to obtain the spatial distribution weight;

[0023] The second clustering module is used to add spatial distribution weights to the distance metric of the WKmeans algorithm and perform KMeans clustering to obtain weighted clustering results;

[0024] An iterative update module is used to iteratively update the cluster center, the category to which the sample point belongs, and the feature weight based on the weighted clustering result until the weighted clustering result converges or reaches a preset number of iterations;

[0025] The judgment module is used to evaluate the weighted clustering result to obtain an evaluation result, and if the evaluation result does not meet the preset requirements, execute the first clustering module, the calculation module, the second clustering module and the iterative update module until the preset requirements are met.

[0026] In a possible implementation, the second clustering module includes:

[0027] Generate a submodule for adding spatial distribution weights to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance;

[0028] The KMeans clustering submodule is used to perform KMeans clustering using weighted Euclidean distance to obtain weighted clustering results.

[0029] In a possible implementation, the preprocessing includes data cleaning to eliminate noise effects, feature transformation, normalization, and standardization.

[0030] In a possible implementation, the spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

[0031] In a possible implementation, the estimated K value is used to provide initial cluster number and spatial distribution information.

[0032] The above main scheme of the present application and its further options can be freely combined to form multiple schemes, all of which are schemes that can be adopted and claimed for protection in the present application; and in the present application, (non-conflicting options) options and other options can also be freely combined. After understanding the scheme of the present application, those skilled in the art can understand that there are many combinations based on the prior art and common knowledge, all of which are technical schemes to be protected by the present application, and they are not exhaustively listed here.

[0033] The present application discloses a pulse clustering method and device based on WKMeans, which first pre-processes the original pulse to obtain a pre-processed pulse, initializes the DOA parameters of the pre-processed pulse in spatial characteristics and directional characteristics to obtain the estimated K value of the clustering tree diagram, and then calculates the probability density of each data point in the mixed Gaussian distribution GMM to obtain the spatial distribution weight, adds it to the distance metric of the WKmeans algorithm to perform KMeans clustering to obtain a weighted clustering result, iterates and updates the cluster center, the category to which the sample point belongs, and the feature weight until the weighted clustering result converges or reaches a preset number of iterations, evaluates the weighted clustering result to obtain an evaluation result, and repeats the above steps until the preset requirements are met if the preset requirements are not met. By introducing DOA unsupervised hierarchical clustering, PDW weights and parameter normalization spatial distance optimization strategies in the iterative process of KMeans, accurate clustering division of electromagnetic radiation pulse data is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flow chart of a WKMeans-based pulse clustering method proposed in an embodiment of the present application is shown.

[0036] Figure 2 The original data distribution diagram proposed in the embodiment of the present application is shown.

[0037] Figure 3 A distribution diagram of the kmeans clustering results proposed in an embodiment of the present application is shown.

[0038] Figure 4 A distribution diagram of clustering results based on GMM-WKmeans proposed in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0039] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0040] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0041] In the prior art, the K-means clustering algorithm completes clustering by iteratively searching for k center points in a data set and assigning data points to the cluster where the nearest center point is located. When processing data containing multiple features, all features are usually given the same weight. This results in that in some practical applications, features of different importance are treated equally in the clustering process, affecting the accuracy and effectiveness of the clustering effect.

[0042] Therefore, in order to solve the above-mentioned technical problems, the embodiment of the present application proposes a pulse clustering method and device based on WKMeans. On the basis of the KMeans algorithm, the method introduces a spatial distribution weight mechanism. First, the GMM algorithm is used to model the spatial distribution, determine the overall probability distribution of the data, and weight the clustering contribution of different data points by calculating the weights of the maximum probability density function of different sample points, thereby improving the clustering performance of the algorithm. In the iterative process of KMeans, by introducing DOA unsupervised hierarchical clustering, PDW weights and parameter normalization space distance optimization strategies, efficient and accurate clustering division of large-scale electromagnetic radiation pulse data is achieved. It can effectively solve the problems of low computational efficiency and poor clustering effect of the traditional K-means algorithm when facing large-scale radar pulse radiation source data sets, and is particularly suitable for effective clustering of PDW data with different feature weight characteristics, which will be described in detail below.

[0043] Please refer to Figure 1 , Figure 1 The flowchart of a WKMeans-based pulse clustering method proposed in an embodiment of the present application is shown. The method is based on GMM-WKmeans and is an extension and optimization based on the K-means algorithm, and includes the following steps:

[0044] Step S1, preprocessing the original pulse to obtain a preprocessed pulse.

[0045] Preprocessing includes data cleaning to remove noise effects, feature transformation, normalization, and standardization.

[0046] The parameters of the original pulse are preprocessed, in which data cleaning removes non-interesting components in the signal, such as random noise, periodic interference or outliers, which can retain the main characteristics of the signal while reducing or eliminating the negative impact of noise. Feature transformation converts the original signal into a set of new features that are more meaningful or easier to analyze through mathematical transformation. Normalization adjusts the scale of the signal so that all data falls into a specific range, such as between 0 and 1, to eliminate the impact of different magnitudes. Standardization converts data into a form with specific statistical properties (usually a mean of 0 and a standard deviation of 1), which is conducive to comparing the importance of different features.

[0047] Step S2: Initialize hierarchical clustering of the DOA parameters of the preprocessed pulses in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram.

[0048] The estimated K value is used to provide the initial cluster number and spatial distribution information.

[0049] Initializing hierarchical clustering will first select a suitable distance metric to measure the similarity or distance between different DOA values, such as Euclidean distance, cosine similarity, etc. When hierarchical clustering starts, each data point (here, the DOA value of each pulse) is regarded as a separate cluster.

[0050] After that, a cluster tree is constructed, and the most similar clusters are gradually merged. Each merger will form a new cluster, and the "distance" required for this merger will be recorded. This process continues until all data points are merged into a large cluster, and a cluster tree (dendrogram) is generated in the process.

[0051] Observe the cluster dendrogram and estimate the optimal number of clusters K (estimate K value) based on the "elbow" principle or specific cutting criteria in the dendrogram. The elbow refers to the point where the increase in the distance between clusters suddenly decreases as the number of clusters increases. This point indicates that increasing the number of clusters after this point has little marginal benefit in reducing internal differences.

[0052] Based on the estimated K value, the data is segmented to obtain the final clustering result, in which pulses with similar DOA values ​​and spatially clustered or continuous distribution characteristics are divided into the same cluster.

[0053] Step S3: Based on the estimated K value, the probability density of each data point in the mixed Gaussian distribution GMM is calculated to obtain the spatial distribution weight.

[0054] First, the K clusters determined set K mixed Gaussian models. Each model is defined by a series of parameters, including the mean (μ) of each Gaussian distribution, the covariance matrix (Σ) and the mixing coefficient (π) of each model. These parameters together determine the probability that a data point belongs to the model.

[0055] The EM (Expectation-Maximization) algorithm is used to iteratively optimize the parameters of the GMM. The EM algorithm is divided into two steps: the E-step (expectation step) estimates the probability that each data point belongs to each Gaussian component, and the M-step (maximization step) re-estimates the model parameters based on these probabilities to maximize the likelihood of the data. This process is repeated until the parameters converge. Once the GMM model parameters converge, the probability density function value of each data point under each Gaussian component is calculated, and then the overall probability density of the point is obtained by weighted summation based on the mixing coefficient. This value reflects the relative importance of the data point in the spatial distribution, that is, the spatial distribution weight W.

[0056] After obtaining the weight W of each data point, further analyze the data, such as emphasizing important data points in the clustering results, or considering these weights in subsequent analysis to more accurately reflect the spatial distribution characteristics and density information of the data.

[0057] Step S4: adding the spatial distribution weight to the distance metric of the WKmeans algorithm and performing KMeans clustering to obtain a weighted clustering result.

[0058] The spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

[0059] Step S4, adding spatial distribution weights to the distance metric of the WKmeans algorithm and performing KMeans clustering to obtain weighted clustering results, includes:

[0060] Step S4-1, adding the spatial distribution weight to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance;

[0061] Step S4-2: Use the weighted Euclidean distance to perform KMeans clustering to obtain a weighted clustering result.

[0062] In the weighted Euclidean distance, each dimension is multiplied by a weight, which is usually adjusted according to the previously obtained GMM probability density value. A larger weight means that the difference of the feature is given more consideration when clustering. Before the KMeans clustering starts, the weighted Euclidean distance is used to modify the distance metric method, which means that when calculating the distance from each data point to each cluster center, the feature weight W of the data point will be considered, which will directly affect the decision process of assigning the data point to the nearest cluster.

[0063] Use the modified distance metric to execute the KMeans algorithm, initialize the cluster center (optionally random or KMeans++ method), and then iterate the subsequent steps until convergence: assign each data point to the nearest cluster center according to the weighted Euclidean distance. Recalculate the center of each cluster, usually taking the weighted average of all points in the cluster (here the weight is the spatial distribution weight W calculated previously), check whether the change in the cluster center is less than the predetermined threshold or reaches the maximum number of iterations, if so, stop, otherwise continue iterating.

[0064] After clustering is completed, the clustering quality is tested using evaluation indicators such as the silhouette coefficient, and the weights or clustering parameters may need to be adjusted based on the evaluation results to optimize the results.

[0065] Step S5: iteratively update the cluster centers, the categories to which the sample points belong, and the feature weights based on the weighted clustering results until the weighted clustering results converge or a preset number of iterations is reached.

[0066] Set the initial cluster center, randomly select or use the KMeans++ method to optimize the selection of the initial center, and then perform an iterative loop to assign each sample point to the cluster where the nearest cluster center is located according to the weighted Euclidean distance. In some variants, the feature weights are dynamically adjusted according to the clustering results, and then the weighted centroid of each cluster is calculated as the new cluster center. The weight here can be uniform or the spatial distribution weight W obtained previously through GMM.

[0067] Determine whether the change of the cluster center is less than a preset threshold or reaches the preset maximum number of iterations. If any condition is met, stop the iteration and output the current clustering result.

[0068] Step S6: Evaluate the weighted clustering result to obtain an evaluation result. If the evaluation result does not meet the preset requirements, repeat steps S2 to S5 until the preset requirements are met.

[0069] After each iteration, the silhouette coefficient is calculated after the clustering result is obtained. In Python's scikit-learn library, the metrics.silhouette_score function is used to pass in the sample feature matrix, cluster label and weighted Euclidean distance to obtain the overall silhouette coefficient and analyze whether the silhouette coefficient meets the preset requirements. If the value is lower than the set threshold, it indicates that the clustering effect is not good and further adjustment is needed. Otherwise, the clustering result is considered to be satisfactory.

[0070] If the evaluation results show that the clustering quality does not meet the requirements, return to the clustering initialization step, adjust the parameters or strategies, and then fine-tune the algorithm parameters, such as adjusting the upper limit of the number of iterations, the distance threshold, etc. Based on the evaluation feedback, continue to iterate the above process, that is, adjust-cluster-evaluate, until the silhouette coefficient of the clustering result or other selected evaluation indicators reaches a satisfactory level.

[0071] In the embodiment of the present application, the WKmeans algorithm achieves more refined and accurate clustering of radar radiation source data in complex scenes by assigning weights to data points and combining the characteristics of hierarchical clustering to automatically determine the number of clusters. Compared with the traditional K-means algorithm, WKmeans shows higher clustering efficiency and accuracy when processing large data sets, and can better explore the potential structure and patterns inherent in the data, thereby providing more powerful support for subsequent data analysis and decision-making. The improved WKmeans clustering algorithm of the present invention can significantly improve clustering performance and result quality.

[0072] In a possible implementation, in order to specifically demonstrate the implementation process of the WKmeans algorithm, a large-scale PDW dataset containing 10,000 samples is selected as an example.

[0073] First, the dataset was subjected to detailed data preprocessing steps, including data cleaning to remove outliers and missing values. Then, an unsupervised hierarchical clustering algorithm was used to hierarchically cluster the DOAs and obtain a cluster tree. The initial number of clusters K was determined by cutting the tree at a specific height.

[0074] Then, a weight assignment method based on mixed Gaussian distribution is used to weight the PDW data, increase the weight of dense area data, model the spatial distribution of data points through the GMM algorithm, and calculate the distribution weight value corresponding to each data point based on the probability density information.

[0075] Then, the weight-optimized data is used to perform K-means clustering operations. The clustering effect is evaluated by calculating the silhouette coefficient, contour coefficient and other indicators of each cluster. If the evaluation result does not meet the requirements, the above weight update and clustering operations are executed repeatedly until the pre-set clustering convergence criteria are met.

[0076] For a radar radiation source data set in a complex scene, each PDW parameter represents a radar target of each type. Traditional K-means cannot identify which bands of data are more important. However, using the improved WKmeans algorithm of the present invention, weights can be assigned according to the distribution of each type of data on the overall data, so that the clustering process can focus more on areas with higher density and representativeness, and provide more refined clustering capabilities for high-density areas, realizing an information clustering algorithm based on WKmean, which has higher clustering performance and robustness, and is suitable for clustering analysis of various large-scale, high-dimensional and dense data sets.

[0077] Based on the above reference Figures 2 to 4 , Figure 2 shows the original data distribution diagram proposed in the embodiment of the present application, Figure 3 The distribution diagram of kmeans clustering results proposed in the embodiment of the present application is shown. Figure 4 A distribution diagram of clustering results based on GMM-WKmeans proposed in an embodiment of the present application is shown. It can be seen that the algorithm can effectively fit and cluster the distribution of the original data.

[0078] A possible implementation of a pulse clustering device based on WKMeans is given below, which is used to execute each execution step and corresponding technical effect of the pulse clustering method shown in the above embodiment and possible implementation. The device includes:

[0079] A preprocessing module, used for preprocessing the original pulse to obtain a preprocessed pulse;

[0080] The first clustering module is used to initialize hierarchical clustering of the DOA parameters of the pre-processed pulses in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram;

[0081] A calculation module is used to calculate the probability density of each data point in the mixed Gaussian distribution GMM based on the estimated K value to obtain the spatial distribution weight;

[0082] The second clustering module is used to add spatial distribution weights to the distance metric of the WKmeans algorithm and perform KMeans clustering to obtain weighted clustering results;

[0083] An iterative update module is used to iteratively update the cluster center, the category to which the sample point belongs, and the feature weight based on the weighted clustering result until the weighted clustering result converges or reaches a preset number of iterations;

[0084] The judgment module is used to evaluate the weighted clustering result to obtain an evaluation result, and if the evaluation result does not meet the preset requirements, execute the first clustering module, the calculation module, the second clustering module and the iterative update module until the preset requirements are met.

[0085] In a possible implementation, the second clustering module includes:

[0086] Generate a submodule for adding spatial distribution weights to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance;

[0087] The KMeans clustering submodule is used to perform KMeans clustering using weighted Euclidean distance to obtain weighted clustering results.

[0088] In a possible implementation, the preprocessing includes data cleaning to eliminate noise effects, feature transformation, normalization, and standardization.

[0089] In a possible implementation, the spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

[0090] In a possible implementation, the estimated K value is used to provide initial cluster number and spatial distribution information.

[0091] Therefore, the present application discloses a pulse clustering method and device based on WKMeans, firstly, the original pulse is preprocessed to obtain a preprocessed pulse, the DOA parameters of the preprocessed pulse are initialized hierarchical clustering in spatial characteristics and directional characteristics to obtain the estimated K value of the clustering tree diagram, and then the probability density of each data point in the mixed Gaussian distribution GMM is calculated to obtain the spatial distribution weight, which is added to the distance metric of the WKmeans algorithm to perform KMeans clustering to obtain a weighted clustering result, and the cluster center, the category to which the sample point belongs, and the feature weight are iteratively updated until the weighted clustering result converges or reaches a preset number of iterations, and the weighted clustering result is evaluated to obtain an evaluation result. If the preset requirements are not met, the above steps are repeated until the preset requirements are met. By introducing DOA unsupervised hierarchical clustering, PDW weights and parameter normalization spatial distance optimization strategies in the iterative process of KMeans, accurate clustering division of electromagnetic radiation pulse data is achieved.

[0092] Therefore, compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0093] First, it can better adapt to data sets with heterogeneous feature importance and avoid clustering bias caused by uniform weights;

[0094] Second, by modeling feature weights through mixed probability distribution, clustering pays more attention to dense key target needs and improves the ability to analyze the internal structure of data;

[0095] Third, by analyzing the data features and reasonably allocating the weights of data points, the clustering contribution of important data is improved, thereby improving the accuracy of clustering results while maintaining computational efficiency.

[0096] Fourth, the DOA dendrogram based on hierarchical clustering can automatically determine the number of initial K and provide physical prior information more intuitively.

[0097] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A pulse clustering method based on WKMeans, characterized in that: The method comprises: Step S1, preprocessing the original pulse to obtain a preprocessed pulse; Step S2, performing initial hierarchical clustering on the DOA parameters of the preprocessed pulse in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram; Step S3, based on the estimated K value, calculate the probability density of each data point in the mixed Gaussian distribution GMM to obtain the spatial distribution weight; Step S4-1, adding the spatial distribution weight to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance; Step S4-2, using weighted Euclidean distance to perform KMeans clustering to obtain weighted clustering results; Step S5, iteratively updating the cluster center, the category to which the sample point belongs, and the feature weight based on the weighted clustering result until the weighted clustering result converges or reaches a preset number of iterations; Step S6: Evaluate the weighted clustering result to obtain an evaluation result. If the evaluation result does not meet the preset requirements, repeat steps S2 to S5 until the preset requirements are met.

2. The pulse clustering method according to claim 1, characterized in that: The preprocessing includes data cleaning to eliminate noise effects, feature transformation, normalization and standardization.

3. The pulse clustering method according to claim 1, characterized in that: The spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

4. The pulse clustering method according to claim 1, characterized in that: The estimated K value is used to provide initial cluster number and spatial distribution information.

5. A pulse clustering device based on WKMeans, characterized in that: The device comprises: A preprocessing module, used for preprocessing the original pulse to obtain a preprocessed pulse; The first clustering module is used to initialize hierarchical clustering of the DOA parameters of the pre-processed pulses in terms of spatial characteristics and directional characteristics to obtain an estimated K value of the clustering dendrogram; A calculation module is used to calculate the probability density of each data point in the mixed Gaussian distribution GMM based on the estimated K value to obtain the spatial distribution weight; Generate a submodule for adding spatial distribution weights to the distance metric in the WKmeans algorithm to obtain a weighted Euclidean distance; KMeans clustering submodule, used to perform KMeans clustering using weighted Euclidean distance to obtain weighted clustering results; An iterative update module is used to iteratively update the cluster center, the category to which the sample point belongs, and the feature weight based on the weighted clustering result until the weighted clustering result converges or reaches a preset number of iterations; The judgment module is used to evaluate the weighted clustering results to obtain evaluation results. If the evaluation results do not meet the preset requirements, the first clustering module, the calculation module, the generation submodule, the KMeans clustering submodule and the iterative update module are executed until the preset requirements are met.

6. The pulse clustering device according to claim 5, characterized in that: The preprocessing includes data cleaning to eliminate noise effects, feature transformation, normalization and standardization.

7. The pulse clustering device according to claim 5, characterized in that: The spatial distribution weight is determined according to the spatial distribution and density characteristics of the data points.

8. The pulse clustering device according to claim 5, characterized in that: The estimated K value is used to provide initial cluster number and spatial distribution information.

Images