Research On Time Series Classification Algorithm Based On Residual Network

Time series data exists widely in our daily life.The time series classification problem is one of the important problems facing.Existing time series classification methods can be roughly divided into distance-based methods,feature-based methods and deep learning-based Methods,traditional methods require manual processing of features and artificial selection of classifiers.Some deep learning methods are end-to-end methods,which show good classification results,but there is no effective design of the network structure for the characteristics of time series data.Time series data often faces two problems: the choice of time scale and noise interference.A robust time series classification algorithm should be able to capture time series data at different time scales,because long-term features reflect the overall trend and short-term features reflect subtle changes in local areas.At the same time,time series data is easily disturbed by noise and loses its meaning.Whether effective time scale selection and effective noise processing can be performed on time series data will have an important impact on the time series classification effect.At present,existing time series classification algorithms that take into account the time scale are not ideal,and rarely can The noise problem proposes targeted solutions.Based on the above problems,this paper researches on deep learning-based time series classification algorithms,and improves the residual network on the basis of considering time scale and noise to improve the classification accuracy of time series.The main work includes:(1)A time series classification algorithm based on multi-scale residual network is proposed to deal with the time scale selection problem in time series.The structure is mainly divided into a data preprocessing stage including data identity mapping,data smoothing filtering and data sampling,a local convolution stage including convolution operations and pooling operations,three convolution blocks and three residuals Block residual network stage.In order to evaluate the performance of this method,this paper conducted experiments on 85 public data sets of UCR.Experiments show that the method proposed in this paper has better performance than other methods.(2)A time series classification algorithm based on residual contraction network is proposed.In this paper,three different residual shrinkage networks are proposed to deal with different types of data sets for the noise problem in time series data-a residual shrinkage network based on a fully connected network to obtain a threshold,and a residual shrinkage network based on a residual network to obtain a threshold,Based on a fully convolutional network to obtain a threshold residual shrinkage network.Experiments show that the method in this paper has obvious improvement on the classification effect of time series data compared with the traditional residual contraction network,and it can effectively improve the classification accuracy compared with the existing time series classification algorithm.