Font Size: a A A

Time Series Sign Aggregation Approximation Combined With Fluctuation Rate And Its Application

Posted on:2021-09-24Degree:MasterType:Thesis
Country:ChinaCandidate:L WangFull Text:PDF
GTID:2480306032967089Subject:Computer technology
Abstract/Summary:PDF Full Text Request
In recent years,with the rapid development of information technology,the efficiency and intelligence of various machines and equipment have been continuously improved,and more and more time series data have been collected in people's daily life and work.In order to effectively mine and analyze the hidden information and potential value of these data,many scholars have proposed many targeted methods.However,because time series data itself has the characteristics of high dimensionality and large amount of data,the analysis and mining of it directly has high computational complexity and low efficiency,so time series data usually needs to be reduced in dimension.Traditional dimensionality reduction methods are difficult to effectively capture the morphological characteristics of time series data.To solve this problem,this paper proposes a time series symbol aggregation approximation method that incorporates volatility and applies it to time series data classification and fault diagnosis problems.The main research contents are as follows:(1)Approximate representation of time series symbol aggregation that incorporates volatility:In this paper,the volatility index is defined on the basis of traditional symbolization methods to simultaneously quantify the volatility amplitude and change trend information of the time series to compensate for the loss of information after dimensionality reduction Problem;then use the symbol vector of fused volatility to approximate the subsequence.(2)Time series similarity calculation and classification:A new time series distance measurement method is given on the basis of fusion of time series symbol aggregation approximation of volatility.Based on this measurement method,a similarity calculation and classification method for time series is proposed,and classification learning experiments are conducted on the data sets in 20 UCR time series classification archives to verify the effectiveness of the proposed method.(3)Time series symbol aggregation approximation and fault diagnosis:On the bearing fault data set of Case Western Reserve University,combined with the superiority of CNN and LSTM,this paper proposes a bearing fault diagnosis model.Based on the time series symbol aggregation approximation method of fusion volatility,the original time series is converted into a volatility sequence and a symbol mean sequence,and then these two sequences are respectively input to the CNN module and the LSTM module in order to diagnose the bearing malfunction.Combined with UCR time series classification archives and Case Western Reserve University(CWRU)rolling bearing fault diagnosis data,the proposed scheme was evaluated experimentally.From the experimental results,it can be seen that the time series symbol aggregation approximation method of fusion volatility has obtained better classification accuracy and time efficiency than the traditional symbol aggregation approximation method on most data sets.In terms of fault diagnosis,an improved method using time series symbol aggregation approximation of fused volatility reduces the accuracy of diagnosis after the original data is reduced in dimension,but the time efficiency is greatly improved.
Keywords/Search Tags:Time Series Analysis, Symbolic Aggregation Approximation, Classification Learning, Fault Diagnosis, Deep Learning
PDF Full Text Request
Related items