Univariate time sequence classification method based on graph neural network

Abstract

The invention discloses a graph neural network-based univariate time sequence classification method. The method comprises the following steps of preliminarily processing univariate time sequence databy using a plurality of one-dimensional convolution layers with different convolution kernel sizes, sparsifying an obtained feature vector by using a ReLU activation function, and constructing a weighted undirected network graph by using the obtained feature vector according to a certain rule; and finally, classifying the obtained network graphs by using a network graph classification model in thegraph neural network field so as to realize a classification task of a univariate time sequence data set. The self-learning ability of the neural network is utilized to enable the time sequence to belearnably mapped into a suitable network graph; in the mapping process, the feature information contained in the original time sequence is fully reserved, compared with a visual graph networking method for fixedly converting the time sequence into a network graph, operation complexity is reduced, the processing speed is increased, classification precision is improved, and classification accuracyis improved on a single-variable time sequence data set of the Born epilepsy electroencephalogram; classification precision achieved by the method is superior to the precision obtained by a visual graph networking method, and the processing speed is remarkably improved.

Description

technical field [0001] The invention relates to network science, data mining and data analysis technology, in particular to a univariate time series classification method based on a graph neural network. Background technique [0002] A time series is a sequence containing important information composed of data points in the time domain, in which there is often the same time interval between adjacent data points on the time level. At present, general time series data can be divided into univariate time series and multivariate time series. The elements in univariate time series come from the measured value of the same variable, that is, one time point corresponds to only one value, while the elements of multivariate time series Each time point corresponds to multiple values. Time series can be seen everywhere in life, and its applications are quite extensive, such as weather forecasting, financial forecasting, ECG classification, EEG classification and radio modulation signal...

AI Technical Summary

Problems solved by technology

In addition, with the development of the complex network field, some people map the time series into a network graph through some visual graph network building methods, and then use the network graph classification method to classify the network graph, and finally realize the goal of classifying the time series. However, the current mapping method for obtaining network diagrams is relatively rigid, and cannot give full play to the advantages of network diagrams, and the mapping process is complicated and takes a lot of time

[0005] At present, the typical method of mapping the time series into a network graph and then classifying it is not uncommon, but most of the mapping methods are not flexible, and the whole classification process is relatively complicated

Images