Evolutionary clustering method for time series data with heterogeneous features based on graphics processing unit

Abstract

A graphical processing unit-based time-series data evolution clustering method for heterogeneous features, including the following steps: (1) Extracting original data features, using multi-view to represent the heterogeneous features of the original data; (2) Applying for video memory space, and using The data transfer function provided by the graphics processing unit transfers the data to the video memory of the graphics processing unit; (3) performs multi-matrix non-negative decomposition on the graphics processing unit, and iteratively updates the characteristic modulus matrix, timing modulus matrix and data object modulus matrix, Until the objective function converges; (4) Normalize the modulus matrix to obtain the membership probability of each feature in the cluster, the time series evolution trend of the cluster and the probability that each data object belongs to each cluster; (5) Finally, release the memory space . The invention utilizes the high concurrency of the graphics processing unit to accelerate the multi-matrix non-negative decomposition process, introduces a time-series feature view into the multi-view representation, and uses the time-series modulus matrix after the multi-matrix non-negative decomposition to obtain the evolution trend of clusters over time.

Description

Technical field [0001] The invention relates to a parallel heterogeneous feature time series data evolution clustering method, in particular to a heterogeneous feature time series data evolution clustering method based on a graphics processing unit, and belongs to the technical field of data processing. Background technique [0002] In the real world, most data have time characteristics, such as social media data, stock data, medical data, scientific literature data, etc. The time features in these data can be used to discover the evolution trend of events, detect abnormal behaviors, and predict event development. Time series data evolutionary clustering has many potential needs in real applications. For example, in social media data, a large number of users publish, communicate, disseminate and track various hot events related to society and politics. This process reflects people in real time. Views and opinions on the event being spread. Therefore, for government departments,...

AI Technical Summary

Problems solved by technology

[0007] The technical problem solved by the present invention is to propose a method for evolutionary clustering of time-series data with heterogeneous features based on graphics processing units, which overcomes the inability of the prior art to effectively use heterogeneous features in data for evolutionary clustering and the large amount of data. The problem of slow calculation speed

Images