Sequential Pattern Mining Face To Time Series Similarity And Application

With the gradual expansion of the data mining research, time series data analysis techniques has been frequently used in various areas of finance, transportation, environment, medicine and other fields and it is becoming increasingly important for us. Traditional time series data analysis technique is mostly focuses on setting up the various mathematical model for the whole period of time sequence and neglect the importance of a specific interval of time series analysis for the results; on the other hand, for the traditional time series data analysis technology, the research object is mostly focus on the single variable time series and not considering the effects of multiple factors for the multivariate time series; in addition, the research of the time sequence pattern mining method is directly symbolized a large number of historical time series, it has the problem of low pertinence and slower efficiency.For the above shortcomings, based on the traditional time series similarity search algorithm and classic time sequence pattern mining algorithm, in the aspect of single variable time series, studied the time sequence of weights allocation algorithm and the single variable weighted time series similarity search algorithm; in the aspect of multivariate time series, this paper puts forward a time series weights allocation algorithm for different variables and multivariable weighted time series searching algorithm; on this basis, optimized the time sequence pattern mining process; in addition, the actual transaction price time series A shares in Shanghai and Shenzhen are as the experimental data, the experiments have been carried out to verify the research achievements of this paper. Main work and innovations are as follows:(1) Put forward the single variable time series allocation algorithm for each time sequence and the single variable weighted time series similarity search algorithm. For each interval time series has different properties and different effectiveness for search results, based on the term correlation calculation method in information searching area, the time sequence interval weights allocation algorithm is proposed and introducing the weights to the traditional cosine similarity measure method for weighted univariate time series similarity search. It is verified by experiment that the accuracy of the weighted univariate time series search results is higher than the original methods;(2) This paper proposed a multivariate time series similarity search weights allocation algorithm for each variable and multivariate weighted time series similarity search algorithm. In response to the lack of weighted univariate time series similarity search algorithms only consider the individual factors and reference to relative entropy method of multiple attribute decision making problems, designed a variable weights allocation algorithm which is suitable for the multivariate time series similarity search method and the corresponding multivariate time series similarity search algorithm. By the comparing experiment, it is concluded that for the proposed algorithm, the similarity between fol ow-up sequence and the ideal share price sequence is higher.(3) This paper designed a time sequence pattern mining scheme based on the weighted time series similarity search. For the existing time sequence pattern mining method has been targeted poor, time consumption and has other shortcomings, combined with the weighted multivariate time series similarity search algorithm and differential symbolic method to optimize the time sequence pattern mining pretreatment process, so that the next sequential pattern mining symbol set is more targeted and with small dimension. Through the experiment, it conclude that the finding maximum frequent item can be applied directly to the historical time series of symbols, the experimental results shows that the search results has been greatly improve the efficiency of the time sequence pattern mining and also further enhance the accuracy of the mining.