面向车载设备数据流的异常检测方法

Anomaly detection method for on-board equipments data flow

针对现有异常检测方法难以通过车辆行驶数据有效地发现车载设备异常的问题,提出一种面向车载设备数据流的异常检测方法。首先,从稳定性、完整性和一致性三种角度分别计算数据的波动、缺失和差异程度,并将其作为检测值放入累计池。然后,采用改进的Dempster-Shafer证据理论提取累计池中若干检测值的多个异常特征,并合成特征值,当特征值达到阈值时触发为相应异常事件。最后,结合贝叶斯理论建立概率Petri网模型,通过若干异常事件的相互组合推导出设备异常。实验结果表明,在分类模型评价指标下该方法的F均值达到近84%,能够有效地检出可能发生异常的设备。

Aiming at the problem that the existing anomaly detection methods are difficult to find the anomaly of on-board equipments effectively through the driving data of vehicles,this paper proposes an abnormal detection method for on-board equipments data flow.First,the method calculates the fluctuation,deletion and difference of data from the perspectives of stability,integrity and consistency,and then puts the calculated results into the accumulation pool as the detection value.After extracting multiple abnormal features from several detection values in the cumulative pool,the improved Dempster Shafer evidence theory is used to synthesize multiple features into eigenvalues.When the eigenvalues reach the threshold,the corresponding abnormal events could be triggered.Finally,Bayesian theory and probabilistic Petri nets are combined to model the combination of abnormal events and deduce the fault.Experiments show that the classification model evaluation value F mean of this method reaches nearly 84%,which can effectively detect the equipment that may have abnormalities.