Research On Multivariate Time Series Classification By Using Gaussian Model And Neural Network

Multivariate time series(MTS)classification is the process of extracting features or training models based on the known MTS to label the unknown MTS with the known labels.This work is one of the most challenging problems in the field of data mining and pattern recognition.In the research of MTS classification,the traditional methods classify by extracting features or by calculating distance.However,at the age of big data,a large number of time series data is generated,involving all aspects of human production and life,such as financial fields,medical and health fields,industrial production,human pattern recognition,etc.The diversity and complexity of the data is increasing,making MTS classification more difficult.Therefore,the study of MTS classification has important theoretical significance and broad application prospects.The research work in this dissertation is as follows:(1)There is a lot of correlated information between the variables in the MTS.Therefore,we use the parameters of the multivariate Gaussian model to characterize the relationship among them.The important parameter is the covariance can be used to identify and capture the correlated information between the variables.The MTS is converted to the multivariate Gaussian model parameters for modeling,then the multivariate Gaussian model is used to characterize the MTS.The Kullback-Leibler(KL)divergence between multivariate Gaussian models is derived as a similarity measurement to implement the classification.The experimental results verify that the method based on KL divergence and multivariate Gaussian model is superior to the comparative algorithms on accuracy,which confirms the effectiveness of the method.(2)By fully analyzing MTS data,it has the sequential properties and the characteristics of correlation between the variables,just like image data.We apply FCN to implement the MTS classification,which works well in computer vision field.The FCN-based MTS classification method can automatically extract features and reduce the workload of manually designing models.The experimental results show that the method is effective.And the method could combine with the multivariate Gaussian model,using the model parameters instead of the original MTS to train FCN can reduce the amount of calculation and greatly increase the training speed.