Convolutional neural network-based multivariate time series data classification method

Abstract

The invention discloses a convolutional neural network-based multivariate time series data classification method. The method comprises the steps of S1: obtaining multivariate time series data; S2: performing denoising preprocessing on the obtained multivariate time series data; S3: performing dimension reduction on the preprocessed multivariate time series data by adopting a convolutional neural network; and S4: performing segmentation on the data obtained by the dimension reduction by adopting a segmentation aggregation algorithm, calculating a Euclidean distance of the aggregated series data, and according to the Euclidean distance, defining a threshold for performing distinguishing and forming a classification result. The method has the beneficial effects that basic structure features of original multivariate time series data can be better reserved and can be subjected to classification analysis by adopting the segmentation aggregation method; the original multivariate time series data is subjected to dimension reduction representation by adopting the convolutional neural network; the result after the dimension reduction representation is subjected to feature extraction by adopting the segmentation aggregation method; and finally the result subjected to the feature extraction forms the classification method by adopting the Euclidean distance.

Description

technical field [0001] The invention relates to the technical field of mining multivariate time series data, in particular to a classification method for multivariate time series data based on a convolutional neural network. Background technique [0002] Multivariate time-series data widely exists in the social industrial production process, especially for the data generated by complex industrial production processes, due to its high dimensionality, large amount of data and often containing noise, the analysis cost of ordinary data mining algorithms is huge, so the multivariate time-series Data dimensionality reduction analysis is a research hotspot in recent years. [0003] The classification research of multivariate time-series data is helpful to the modeling and analysis of complex systems. In practical engineering applications, due to the correlation and interaction between multivariate variables and the huge amount of data, the classification and analysis of multivariat...

AI Technical Summary

Problems solved by technology

An important disadvantage of these methods is that they ignore the structural information of the data itself

In the process of dimensionality reduction, many important information will inevitably be lost, so that the results of subsequent data mining and analysis cannot well reflect the characteristics of the original data.

[0004] Disadvantages: Most of the current multivariate time-series data dimensionality reduction classification methods are based on statistical principal component analysis methods or their improved methods, which are difficult to better reflect the inherent structural characteristics of multivariate time-series data, thus affecting the classification results

Images