Research On Pattern Matching Problem Of Multivariate Time Series

With the development of information technology, time series data mining is becoming thehotspot of the researchers. The initial work of time series data mining techniques is similarityquery, and pattern matching is a basic work of the similarity queries, so pattern matchingoccupies the basic and core status. Because multivariate time series has a higher dimension,existing multivariate time series pattern matching methods have high computationalcomplexity. If we can first reduce the high dimensionality of the multivariate time series, andthen measure them with pattern matching method s and we will solve the problem of thecomputational complexity which high dimensionality leads to. This article improves theprincipal component analysis and proposes the method of the common principal componentanalysis.Then we use Ping An Bank shares to do some simulations to test the effectiveness ofthis method. Combining with the trends distance pattern matching method, we propose aneffective on multivariate time series pattern matching method on the basis of reducing thedimension of the multivariate time series data by using common principal component analysis.The main contents are as follows:1）Due to the high dimension of multivariate time series, the exisiting multivariate timeseries pattern matching methods have high computational complexity when it is used tomeasure the similarity and the principal component analysis method is the most commonlyused method for dimensionality reduction. According to the characteristics of principalcomponent analysis, we propose the method of common principal component analysis, whichis based on principal component analysis method and improve it. The paper introduces theMatlab software to do some simulations on200groups of stock data of Ping An Bank fromMarch1,2013to December27from which chooses six attribute values showingcharacteristics of stock, namely, opening price, the highest and the lowest price, closing price,trading volume and turnover, and does some comparison with the principal componentanalysis.The results showed that the common principal component analysis method can betterreduce the dimension of the original mutidimensional time series compared with the principalcomponent analysis.2）In-depth analysis of the strengths and weaknesses of existing multivariate time series pattern matching method, we introduce the trend distance pattern matching method, and putforward an effective method of pattern marching for multivariate time series which combinedwith dimension reduction method. To reduce the computational complexity in the process ofmeasurement, this paper introduces the common principal component analysis and usesMatlab software to reduce dimension of the200stock data of Ping An Bank fro m March1,2012to December21firstly, and then do some pattern matching on the data afterdimensionality reduction by using the trend distance pattern matching method and verify theeffectiveness of the method by doing contrast experiments with the Dynamic Time Warpingmethod and Singular Value Decomposition method. Experimental results show that thismethod can effectively measure the similarity of multivariate time series data.