自适应安全分析和监测系统机理分析  

Mechanism of the adaptive safety analysis and monitoring system

作　　者：李剑峰[1,2] 何卫峰[2,3] 刘茂[2] 

机构地区：[1]北京市劳动保护科学研究所,北京100054 [2]南开大学城市公共安全中心,天津300071 [3]军事经济学院勤务指挥系,湖北武汉430035

出　　处：《兰州大学学报（自然科学版）》2008年第5期89-93,共5页Journal of Lanzhou University(Natural Sciences)

基　　金："十一五"科技支撑计划(200603746006)资助.

摘　　要：对探测具有低信噪比特征的恐怖分子活动问题,其中包括"恶性的"或("恐怖分子活动")交易和嘈杂的("良性的")交易,提出了以隐马尔可夫模型和动态贝叶斯网络为基础的自适应安全分析和监测系统,用于帮助情报人员识别恐怖威胁,预测恐怖分子的可能活动.作为一种以交易为基础的分析工具,类似于有噪声背景的目标跟踪问题,将一系列在现实交易中可疑的人、地方和事情作为数据输入,利用隐马尔可夫模型,在前置变量法和时序似然比探测算法的基础上,提供描述恐怖分子网络节点状态的部分(不完整)观测数据的软性证据;构建分层贝叶斯网络,软性证据以可能性信息形式从隐马尔可夫模型传送到贝叶斯网络进行信息融合,用于升级贝叶斯网络的推理.和反恐网络模型相结合,自适应安全分析和监测系统收集、共享、识别信息,并且评价和预测恐怖分子网络的状态,为抑制可能的恐怖威胁给出可行性建议.The problem of detecting, tracking, and counteracting terrorist networks was resolved by the adaptive safety analysis and monitoring system based on hidden Markov models and dynamic Bayesian networks for assisting intelligence analysts to identify terrorist threats and predict possible terrorist actions. The basic premise was that terrorist networks can be evaluated using transaction based models. Analogous to the target tracking problem where states （location, velocity, etc.） are observed through noisy measurements, the true states of terrorist activities are detected against a background of noise transactions data （people, places, things） that appear to be suspicious by using hidden Markov models based on forward variable and log likelihood ratio algorithm and by providing local assessments as outputs that are transformed into soft evidence. Bayesian networks were maintained by higher level （decision making） agencies functioning as fusion centers, and by pooling the summarized information in the form of soft evidence to support belief updating. Combined with the counter-terrorism network models, the adaptive safety analysis and monitoring system integrates, shares and identifies information, in order to evaluates and predicts the terrorist network states. It can thus provide early warnings to facilitate preemption and/or support strategic decision making.

关 键 词：隐马尔可夫模型 贝叶斯网络 软证据 Page检验 反恐 

分 类 号：X913.1[环境科学与工程—安全科学]