Reasearch On Data Stream Prediction Management Technologies

A data stream, as a new kind of data form, exists in many information systems. In these systems, it is very important to predict future tendency of data streams. To the best of the knowledge, recent works are not date in the literature on integrative predictive management models over data streams. An integrative predictive management model over data streams is first proposed and relative techniques are studied in this paper.Firstly, based on a data steam management system, the integrative data stream predictive management model includes data stream preprocessor, data stream storage and data stream query processor. The model supports data stream predictive query as well as normal query.Secondly, on the basis of GEP algorithm, this paper designs data stream predictive algorithm with high efficiency and adaptability to produce predictive function. Standard GEP algorithm includes population initialization, genetic population’s selection and individuals’ evaluation. This paper improves the above three steps. The probability-based strategy is proposed to determine the initial population, making the population of genes tend to diversify. Considering the recessive and dominant Hamming distance between the individual, a strategy is designed to choose the genetic information, which can not only ensure the genetic diversity of populations, but also avoid local optimization. In accordance with the fitness of each gene in each location, the mutation probability of each gene in each location is determined, which guides the direction of genetic variation, thereby improving the efficiency of the evolution of the population.Finally, the model implement techniques include:(1) Biased sampling method based on double timestamps to preprocess data streams;(2) HSDI-Tree, which is designed to quickly access historical data to adjust predictive function;(3) Predictive query operator is designed on the basis of CQL;(4) Experiments on stock data sets are conducted. The results show that the predictive error of the model decreases with the scale of training data increasing, and the improved strategies speed up the search of the GEP algorithm.