Research On Complex Event Processing In Probabilistic Stream

With wide application of EDGE (Electronic Data Gathering Equipment) such as RFID (Radio Frequent Identification) and sensor, massive event-type data are generated in our world. Therefore, complex event processing becomes an extremely important research field for these event-type data. Nowadays Complex Event Processing is widely applicable to many fields, such as business activity detecting and predicting, supply chain management, climatic environment monitoring, medical care, and so on. However, most of current complex event processing algorithms are oriented certain data instead of uncertain data and lack the ability of prediction before events occur.Aimed at the above problems, this paper proposes a scheme of Complex Event Processing in probabilistic stream, which includes complex event detection algorithm HSF-CED (Heuristic Search and Filter-Complex Event Detection) achieved by heuristic search and filter, and event prediction algorithm SVC-SVREP (Semantic Vector Clustering-Support Vector Regression for Event Prediction) achieved by semantic vector clustering and SVR modeling. Contributions of this paper mainly reside in:First, for probabilistic stream generated by uncertain events, a kind of probabilistic stream model is constructed, including event model and Bayesian network to express uncertain events detected by EDGE and infer the probability distribution of them. In this process, Conditional Probability Tree (CP-Tree) structure which is used to store conditional probabilities of Bayesian network is proposed, saving storage space and query time compared with traditional Conditional Probability Table.Second, a new complex event detection algorithm HSF-CED in probabilistic stream is proposed in this paper. This algorithm can detect complex events satisfying query requirements from probabilistic stream with a heuristic method, based on a kind of structure called Chain Instance Queues. To improve the efficiency, it achieves lossless filter for the composite events by magnifying probabilistic threshold. And it further filters composite events by setting proper deviation permit, insuring high recall meanwhile improving efficiency.Third, semantic vector is designed to improve practical value of prediction, which expresses probabilistic stream segment semantic in a basic window. The structure can convert different probabilistic stream segments into the same vector in favor of building predicting model.Last, a new semantic vector clustering based event predicting algorithm SVC-SVREP in probabilistic stream is presented. It is achieved by semantic vector clustering and SVR modeling. SVC-SVREP can predict the probability of the target complex event before this event occurs.In addition, the scheme of complex event processing is able to set parameter according requirements. Experiment results shows the methods in this paper can efficiently detect and predict the target complex event in probabilistic stream, and ensure high accuracy at the same time.