Multi-source atmospheric data clustering method based on distribution density

Abstract

The invention discloses a multi-source atmospheric data clustering method based on distribution density. The method comprises the following steps: firstly constructing a data set DS which is composedof M-dimensional data and is N in data size, and judging the clustering trend of the data set DS; secondly, generating a full neighborhood distribution density matrix DDM of a distance matrix DM of the data set; then, taking a distribution density threshold ddth as a parameter, and dividing a density peak value and discrete points of the full neighborhood distribution density matrix DDM; and finally, intercepting an edge matrix E of all the data, and merging part of discrete point into the density peak value to obtain a clustering result. The clustering result is controlled only by using a single parameter of the distribution density threshold, and data with any distribution shape and distribution uniformity can be clustered; and noisy points can be automatically separated.

Description

technical field [0001] The invention belongs to the field of data mining, and in particular relates to a multi-source atmospheric data clustering method based on distribution density. Background technique [0002] In the practical application of big data mining and analysis, data is collected from different sources in different fields or acquired from different feature collectors. For example, a certain image shared on a website often has text tags and descriptions from different sources; a specific news is reported by multiple news organizations; the same semantics (such as hello) is represented in multiple languages; images are described by different types of features . All of these are called multi-source data (or multi-view data). These data show heterogeneity, yet potential associations. In other words, each individual source (or view) in these data has its specific properties for the knowledge discovery task, while different sources usually contain complementary inf...

AI Technical Summary

Problems solved by technology

After continuous development and mutual reference and fusion of the above-mentioned various clustering algorithms, a classic density peak clustering algorithm DPC emerged. It is relatively larger, and thus achieves efficient clustering of data with a single truncated distance parameter to control arbitrary distribution shapes, but not all data sets can accurately find the cluster center through the decision map and the algorithm cannot distinguish noise points, so there are The researchers improved the DPC algorithm, trying to solve the two major problems of determining the cut-off distance of the algorithm and selecting the cluster center. Although t

Images