基于TBM双层融合架构的航路属性异常检测

航路飞行目标的属性异常检测是确保及时发现飞行异常的关键问题.常用的概率框架需要受到先验信息的局限.可传递置信模型（Transferable Belief Model，TBM）不需要先验信息，能高效处理异质信息，但是传统的TBM无法处理时间上的不连续与不确定性，因此针对异常航路目标检测问题，将马尔可夫模型与TBM框架结合，建立了基于TBM的双层融合架构，实现了多特征融合航路属性异常检测.第一层是通过对多属性冲突信息的分析，实现对多特征的检测，并通过特征贡献度分析，对多特征信息进行打折后再融合；第二层是通过在时间序列上的指派融合，对比预测值和观测值差异，检测航路目标异常变化.仿真试验验证，在切换航路场景与偏离回归场景中，相较动态证据推理方法，本文方法具有更好的决策准确性与时间精确度.

The Anomaly Detection Based on TBM Two-Level Fusion Architecture

The track anomaly detection is the key issue to make sure flying anomaly detected in time for the route flight.Traditional probabilistic frameworks are always based on prior probabilities.Transferable belief model (TBM) theory can generalizes the Bayesian approach without prior probabilities and efficiently deal with heterogeneous data.However,the traditional TBM cannot deal with the discontinuity and uncertainty about the time.Considering the existence of unreliable evidence sources,an alternative anomaly detection method is proposed in the framework of transferable belief model (TBM) theory.A two-level architecture fusion system based on TBM is developed.The novelty of this work is that it can detect both unreliable evidence source and abnormal behavior of the targets within our architecture by using a temporal analysis and a new discounting coefficient through introducing the concept of contribution degrees of features.Detection of abnormal behavior is based on a prediction/observation process and the influence of the faulty sources is weakened through discounting coefficients.The simulations show the better accuracy of decision and precision of time compared with the dynamic evidence reasoning method.