Study On Time Series Partial Periodic Patterns Mining Based On Convolution Algorithm

With the rapid development and the wide application of the Computer technologies and the Database technologies, there are a number of data in lots of fields.In the real world, data is arranged in time order. The series, which are composed by this type of data, is called time series. The repeated periodic behavior is very common in time series. The aim of the mining techniques about periodic patterns in time series is to obtain the valuable and implied knowledge and patterns from lots of data for decision-making. Time Series Data Mining has attracted much more attention.This dissertation modifies the conception of convolution, and uses it in our time series partial periodic patterns mining algorithm to mining frequent periodic patterns. This dissertation also uses the conception of Wavelet Transform to mining the partial periodic in the time series data. Firstly, this dissertation gives a general overview of the status of the time series data mining and periodic pattern mining. Secondly, it introduces the core of the related theories, such as, the conception of the time series and periodic pattern, the conception of convolution,the conception of Wavelet Transform,and how to redescribe data in time series.Thirdly,the dissertation introduces the partial periodic pattern mining algorithm which is based on convolution in detail.By the experiment using the synthetic data and the real data,we illustrate the effectiveness of our proposed algorithm.Also, this dissertation uses the Wavelet Transform to mine the implicit periods .Finally, this dissertation summarizes all the work above,and points the researched tendency in the future.The main innovation points of our work are listed as follows:The partial periodic pattern mining algorithm based on convolution is proposed.In the algorithm, first of all, we discretize and symbolize the time series data, then, we use our algorithm on the symbol sequence to mining periodic patterns.Analysing the result of the experiment using the synthetic data (By the way, we import noisy data in the synthetic data.), we modulate the output of the algorithm in order to reduce the runtime and make the output hddlife pro.We propose the partial periodic pattern mining algorithm based on the Wavelet Transform. We use the function of compression and the function of noise reduction of the Wavelet Transform to mine the implicit periods in the time series data.