Discovery Of Temporal Dependencies In Event Sequences

With the great development of data acquisition and storage technology,time series generated in various fields have more types and larger scales.The information collection of time series has become an important research direction of data mining.In system management,mining hidden patterns in historical event sequences helps management personnel to understand the internal operating status of the system,perform system log management,abnormal detection,and fault tracing.In the aspect of event sequence dependent discovery,predefined time windows are often used to mine simple correlations between events in traditional discoveries of frequent episodes.Besides,in the presence of interleaved dependencies only pairwise patterns can be discovered.Therefore,this paper conducts research work on these issues.The main research contents are as follows:(1)This paper first analyzes and describes event sequences and event dependencies,and discusses frequent episodes and pairwise temporal dependencies based on the interleaved dependencies,which belong to temporal dependencies in an event sequence.Then advantages and disadvantages of these methods about temporal dependencies in an event sequence are summarized.(2)This paper proposes a method for detection of time-lag based pairwise temporal dependency episodes,which introduces the concept of time-lag episode discovery.Adjacent Event Matching Set(AEM)algorithm is adopted in the event matching to avoid setting a time window.The probabilistic statistical model is introduced to handle interleaved temporal dependencies.And then the discovery of time lag is formulated as an optimization problem which can be solved in an iterative way.Compared with Iterative Closest Event(ICE)algorithm,the probability model computed by this means is more effective than the time lag mining by ICE to simulate the actual situation.(3)Based on the discovery of pairwise temporal dependencies,this paper studies temporal dependencies between multiple events and proposes an algorithm for discovery of event chains based on Trie tree.The sorted table is used to mine pairwise temporal dependencies.Then,longer chains are generated by constructing the event chain-based Trie tree.Each event chain is qualified according to the extended event instance set.