Research For Predictive Aggregate Queries Processing Over Data Streams Based On The Sliding Window

In recent years, with the continuous development of financial, sensor network, stock analysis, meteorological monitoring, etc, a unlimited flow of data has caused serious concern in the database fields, so that the data stream management technology as the current research focus.The research according to the characteristics of data stream and application not only broadens the field of database research but also of great academic value and application prospect. This article has done in-depth exploration and research for a number of key technologies about data stream predictive aggregate queries.First, this article introduces the characteristics of data streams, applications and research background and current situation; analysis of some data streams reduction technologies, for example, histograms, random sampling, etc; some data streams prediction models are described, such as regression model, exponential smoothing model, etc; analysis the difference of data streams management systems and traditional database by contrasting approaches, and some typical data streams prototype systems are introduced.Then, characteristics and application of data stream research, and then comparing the existing types of data stream predictive models, the Hidden Markov Models which has used by speech recognition areas are introduced, so that designed a new predictive model base on sliding window-sliding window of Markov predictive model. A new data stream reduction technique called compressed histogram of aggregate feature has proposed, and also accomplished an optimized aggregation queries.And then, the predictive modeling of traditional methods, such as curve fitting, linear regression analysis can only meet the polynomial function, and nonlinear function of the invalidity, so this designed a corresponding treatment method base on the predictive modeling. By aggregating network traffic prediction for the instance of the query processing to introduce the data stream predictive aggregate query processing, such as the parameter initialization, model training, model evaluation, etc. Finally, based on the prediction models and treatment methods, and well-known open source data streams processing engine called Borealis, achieved the predictive aggregation query for data streams. Using data of the network traffic database as experimental data, and the results of several experiments by comparative analysis, so the theory and experiment show that the prediction accuracy in forecasting and prediction efficiency has a significant advantage, by contrast, the existing aggregate data stream query processing methods. Therefore, this approach is a effective improved and expanded for the existing data stream forecasting and query processing techniques.