Multivariate Time Series Classification Method Based on Multi-Timescale Echo State Network

Abstract

The invention discloses a multi-variable time series classification method based on a multi-time-scale echo state network. The steps are as follows: for a multi-variable time series, M jump reserve pools with different time jump connection lengths are used as encoders to generate a A multi-time scale echo state representation; using M convolution and pooling layers and a fully connected layer as a decoder, the convolution and pooling layers are used to learn high-dimensional complex features in multi-time scale echo state representations, fully connected The layer is used to fuse the learned features of different time scales to obtain a multi-time-scale feature of the input time series; a Softmax layer is used as a classifier to generate its classification results according to the multi-time-scale features of the input time series. The negative log-likelihood function is used as the loss function, and the model is trained using the backpropagation algorithm and the gradient optimization method. The classification method of the invention achieves a higher accuracy rate on the multivariate time series classification problem.

Description

technical field [0001] The invention relates to the technical field of time series data mining, in particular to a multivariate time series classification method based on a multi-time scale echo state network. Background technique [0002] In the field of time series data mining technology, multivariate time series classification tasks have been widely used in many fields such as finance, medical treatment, and manufacturing industry. Compared with univariate time series, multivariate time series generally have larger data volume, higher dimensionality and stronger correlation. [0003] For multivariate time series classification problems, the current technologies can be roughly divided into three categories, namely distance-based classification methods, feature-based classification methods and dynamic system-based classification methods. The basic idea of ​​the distance-based classification method is based on some well-designed distance calculation methods (such as dynamic...

AI Technical Summary

Problems solved by technology

However, the existing dynamic system-based methods do not explicitly consider the ubiquitous multi-scale structure in time series data and the multi-scale time dependence contained in it. Therefore, this type of method still has improvement in multivariate time series classification problems. Space

Images