Research On Event Processing Based On Bayesian Network

Recently, with the rapid development of collection equipment for electric data, such as sensor and RFID (Radio Frequency Identification), data for some applications are easily generated. Processing complex event is meaningful for people to gain useful information by tackling data, so it is received great attention. In some applications, such as logistics and supply chain, traffic and health care, people would like to tackle with dynamic information to gain relative and meaningful data. At the same time, people are willing to infer some former or later event. However, current event processing methods on reasoning technology is immature. So it remains to be researched.While Bayesian Network based on probability and statistics is an efficient method for dealing with uncertain events inference, the paper would like to process events associating with it. To apply Bayesian Network inference to event streams processing, the paper is based on the theory of the event streams and Bayesian Network to construct the event Bayesian Network. Then we have methods to do a structure optimization. More importantly, we can do inference computing and predict the later or previous events. The main contributions of this thesis mainly reside in:First, a method is proposed to construct Bayesian Network over dynamic event stream. It classifies and statistics the original event stream, then finds the relationship between events to construct the original structure for event stream.Second, the thesis propose two structural optimization methods. The event network structure achieve optimal performance in the structural complexity and event correlation.Third, forward and reverse inference algorithm on the event network structure are proposed, that is, do event prediction and diagnosis. Under the constraints of time and probability, the methods can find the result or cause event.Finally, we expand the method above, we can compute the time interval of the relatively events to describe the different time granularity of them.The theoretical analysis and experimental evaluations show that the methods of the Bayesian inference for complex event processing are feasible and accurate.