Aggregate Query Processing And Optimization Techniques On Uncertain Data

At present, the management of uncertain data is involved in many important areas, such as:data integration, wireless sensor networks, financial surveillance, trends prediction and mobile objects management. Due to inherent and manmade factors, the uncertainty in these data is inevitable, which is represented by their data uncertainty, existence uncertainty and position uncertainty and so on. Currently, extensive research effort has been given to model and query uncertain data recently. Research directions include modeling uncertainty, query evaluation, indexing. Top-k queries and skyline queries, clustering and mining and so on. However, despite that probability aggregate query is very important in practice, this problem remains unexplored.Traditional aggregate query returns summarized information about objects within a given query range, such as the total number of qualified objects. This type of query is important since users may be interested only in aggregate information instead of specific object. For instance, to monitor traffic volume of a crossroad A in rush hours, query "how many vehicles pass A from8AM to9AM today"First, this paper introduces aggregate query technology on certain data, then extends to uncertain data, and presents the definition of probabilistic aggregate queries. After adding aggregated information of uncertain objects to U-tree index, we propose a new index structure aU-tree which is designed for probability aggregate query. Then through the MBR partition, we propose a pruning technique for single object and multiple objects and present the probability aggregate query algorithm based on aU-tree. In order to improve the efficiency and reduce the computing time, we propose an approximate query algorithm based on sampling method, including single and double sampling method. In the experiments, we test the performance of aU-tree. approximate algorithm and the accuracy of approximate algorithm. The results prove that the performance of approximate algorithm is far better than aU-tree, with accuracy above90%at least.