Research And Implementation Of Relationship-Driven Event Recognition Method

With the rapid development of Internet technology and the strong push of application requirements,the scale of software systems has become larger and the interaction of the system has become more complicated.It has become extremely difficult and unreliable to adjust the system artificially.Therefore,it is necessary to establish a kind of system with the ability of autonomous adjustment,that is,intelligent autonomous system.The intelligent autonomous system can monitor and analyze various abnormal events during system running,then generate and execute control strategies based on the abnormal events.Through above adjustment,system can achieve system goals and provide users with stable functions.Among them,the analyze stage obtains real-time monitoring data of the system and recognizes abnormal events that occur in the system.We can see that the analyze stage is the basis for subsequent generation and execution strategies.If the events cannot be effectively recognized in a timely manner,there will degrade the behavior of the autonomous system,or even crash the system.Therefore,in the adjustment process of autonomous system,the accuracy of event recognition methods is very extremely important.However,the current event recognition methods mainly focus on observing the real-time data of system running.They use the predefined mapping relationships between the system state and the event as the event reasoning knowledge to recognize the event.On the one hand,the changeable environmental states during the system running lead to inaccurate and incomprehensive mapping relationships defined before system running.And the limitations of human cognition make it impossible for domain experts to formulate complete mapping relationships.These two aspects lead to the accuracy of event recognition methods to be improved.On the other hand,it's not enough to only rely on the knowledge of mapping relationships to recognize events.There may be causal relationships between events and events,and these relationships can be used to discover events that are not easily detected directly.Consequently,to solve the problem of low accuracy of recognition methods caused by insufficient reasoning knowledge,we need to enrich event reasoning knowledge,so as to improve the accuracy and comprehensiveness of recognition methods.For the above problems,this paper designs and implements an accurate event recognition method of autonomous system.New knowledge factors are explored for event reasoning,so as to improve the accuracy of event recognition.The main work of this paper consists of the following three parts.Firstly,this paper establishes the event relationship mining and modeling methods.This paper designs the event relationship mining method based on the GSP algorithm.This method mine the potential causal relationships between events from the system running logs.These relationships are as a new event reasoning knowledge factor.And this paper proposes an event relationship modeling method based on Fuzzy Fault Tree(FFT),which uses the tree structure,fuzzy numbers and fuzzy operators to model event causal relationships.FFT is convenient for subsequent conversion to the Bayesian Network(BN)model with stronger analysis capabilities.Secondly,this paper establishes an event recognition method based on BN.This paper designs the conversion method of FFT to BN,including the conversion of structure diagram and calculation of conditional probability,as well as the conditional independence test method to verify the correctness and rationality of the BN.Next this paper uses the BN backward reasoning and forward reasoning to comprehensively recognize events in a probabilistic manner.In addition,three important analysis methods are designed to analyze and judge basic event nodes in BN to reduce the probability of complex events.Thirdly,based on Web application system Book Store and a microservice architecture system Train Ticket,this paper designs experiments to verify the relationship-driven event recognition method implemented in this paper,and prove the effectiveness and accuracy of the method in this paper.