Research On A Method Of Time Series Data Correlation Analysis

In recent years,the research on time series correlation has attracted more and more attention of academic staties from domestic and abroad.Industrial data is a kind of time series with limited data volume,and its sequence length is between tens and hundreds.At present,the research on limited length time series correlation and time lag correlation analysis has mainly been focused on time domain methods,such as linear regression,cross correlation and so on.However,the time domain analysis methods,which has certain limitations and could not fully grasp the information contained in the time series.In order to deal with the limitation of time series analysis method,three different time series correlation methods are discussed to analyze the industrial data from time domain,frequency domain and time frequency domain respectively.The main work is as follows:(1)For time domain analysis,the correlation between industrial data time series is explored by using gray correlation degree.For the analysis of time-lag relationship between different industrial data time series,the calculation method of grey time-lag correlation degree is proposed.On the basis of the grey relational degree,the idea of correlation calculation of the cross-correlation function shift sequence is fully adopted.The grey time-lag correlation is calculated by using relative shift.The results show that the method can be used to calculate the delay correlation of the sequence.(2)For the frequency domain analysis,cross-spectral analysis is performed on the correlation between industrial data.Aiming at the finite length problem of industrial time series,the Welch method with Hamming window is applied to estimate it's spectrum,in order to obtain better accuracy.The comparison carried out among Welch method with Hamming window,period graph method and Bartlett method.The results show that the welch method and Hamming window reduces the spectrum leakage and improves the credibility of the crossspectrum analysis results.(3)For time-frequency domain analysis,the maximal overlapping wavelet transform method(MODWT)as a new method is applied to to analyze the correlation of limited volume industrial data.For the particularity of industrial data,we discuss how to select the wavelet functions and set the level of decomposition in the process of wavelet analysis,in order to evaluate the reliability of the analysis results.Then the correlation between the time series of the Ministry of Industry is analyzed,and the results show that this method is basically consistent with the time domain analysis results.