一种基于TF-HBPN的复杂系统行为分析方法

Method of Behavior Analysis for Complex System Based on Hierarchical Bayesian Petri Net with Time Factor

针对大规模复杂系统建模分析时的状态爆炸问题,提出一种扩展了时间因素的分层贝叶斯Petri网模型(Hierarchical Bayesian Petri Net With Time Factor,TF-HBPN),并基于该模型提出一种递归构建方法和递归溯因行为分析方法。该方法首先对观测到的系统行为构造顶层TF-HBPN模型,进而通过分层递归方法将复杂系统并发行为分析问题进行分解,并通过递归溯因推理和时序分析来计算非观测系统行为及其事件链的发生概率,最后将分析结果与正常动作事件链进行对比,分离出干扰信息。实例分析表明,该方法可对大规模复杂系统行为进行快速建模分析,当观测数据存在干扰和缺失时仍能进行系统行为的分析和现象的溯因,分析结果的可信度较高。与其他基于Petri网的复杂系统分析方法相比,该方法建模难度更低,模型表达更为简洁也更易理解。

For the problem that the status is proned to explosive growth when we try to model and analyse a complex system with huge scale,this paper proposed an Bayesian hierarchical Petri net model with the extending time factor（TF-HBPN）,and based on this model proposed a recursive construction method and recursive abductive behavior analysis method.Firstly,our method creates top-level TF-HBPN according to the observed system＇s behavior and decomposes behavior analysis problem of complex systems through hierarchical recursion.Then it calculates the fault probability of the correct time sequence chain of fault events by recursive abductive reasoning.Finally,it calculates the bayesian probability of the event chain of system＇s behavior obtained by recursive abductive reasoning and time series analysis.The analysis results are compared with the right event chain to separate interference information.The experimental cases show that this method can model and analyze complex fault quickly and still can do system＇s behavior analysis and abductive reasoning with alarm missing.Compared with the general Petri nets,this method has a lower degree of modeling difficulty and is more concise and simple.