Time Series Similarity, Aggregate Top-k Query Algorithms And Applications

Time series data is a series of value which often measured at equal time intervals. With the development of techniques, large time series datasets are being generated in a rich variety of practical applications. Examples include stock market, medical observations, traffic monitoring, environment sensor networks, and archaeological data mining etc. Therefore, data mining and management on massive time series has become a new challenge.This paper studies the problem of aggregate top-k query on time series data, similarity matching on time series data and a time series data management prototype system. The main contributions of this work are as follows:In this paper a new time series similarity measure is introduced. The proposed approach lies on a trend invariant distance that is able to manage distortions in time series data. The approach employs a transformation to map the original time series in a new space which is able to take into account different granularities at the same time. The experiments results demonstrated the efficiency and accuracy of our algorithm.We study the problem of aggregate top-k query on time series data and propose an I/O efficient algorithm. Aggregate top-k query becomes very slow on large time series dataset due to many I/O operations during the query and designing fast and scalable query algorithms still remains a very challenging problem. To solve this problem, our solution framework including two phrase:separate pre-rank time series in dataset and early termination criterion during query. At last we exhibit extensive experiments on both synthetic and real datasets.We design and implement a time series data management prototype system. It provides functions including time series stream data mining and massive time series data management system.