Research And Application Of Time Series Similarity Pattern Mining

Data mining is a process that extracts information from large amount of data. And many related researches on specific-application, such as time database, space database, multimedia database and so on, have been carried out. Among these research fields, time series data mining is a rather complex branch,which is a technique that extracts the most valuable information from large amount of history time series data.Time series similarity- pattern mining is a valuable one that extracts similar rules from time series data. And it has great value to do related research because of its widely-used applications.This dissertation mainly carries out the followed researches:Firstly, through analyses on wavelet transform, a database reduction method on time series using Haar wavelets is summarized, and further data analyses are also carried out. By this way, data will be divided into two parts-low-frequency and high-frequency-by making convolution between original data and low-frequency filter, between original data and high-frequency filter in scale descending order, then sampling interval, and replacing the original data with the last low-frequency part, neglecting of high-frequency part.Secondly, this dissertation discusses the problem on time series similarity-pattern mining from similarity measurement, storage structure, searching integrality. Euclidean Distance formula is applied in similarity search. Sliding window technique is employed in Subsequence Matching. And the idea of Minimum Bounding(hyper)-Rectangle is applied in storage structure. Range Query, All-pairs Query and Nearest Neighbor Searches are all realized in this dissertation.Thirdly, this dissertation employes similarity-pattern, which is achieved through similarity search, to forecast time series data. This is a application of time series similarity-pattern mining. The data source is weather data. The main job is the forecast of five kind of bad weather, such as rainstorm, cold wave and so on.Fourthly, particular analyses about whether making time series data reduction with Haar wavelets can improve higher efficiency for similarity search are carried out.Finally, through the application of above researches in weather data, experiment results prove that such time series data similarity search and forecast theory is a good way for time series similarity-pattern mining.