深度递归网络在物联网系统异常检测中应用研究

Application of deep recursive network in anomaly detection of Internet of Things system

物联网系统采用感知层、传输层和应用层三层体系架构,物联网的三层结构具有时序性,上一层数据异常会链式反应到后续层,传统的物联网异常检测方法无法有效识别数据异常并快速定位异常发生在哪一层。文中提出以深度递归网络对物联网系统异常检测进行建模,感知层、传输层、应用层作为深度网络输出层,深度递归网络通过核函数变换能够提取高阶特征,并且深度递归网络本身的时序特性能够提升异常检测的准确性。实验结果表明,深度递归网络在物联网系统异常检测中能够获得较高的检测准确率。

The Internet of Things system adopts the architecture of perception layer,transmission layer and application layer. The three-layer structure of the Internet of Things is chronological. The former layer of data anomaly will be linked to the following layer. The traditional anomaly detection method of the Internet of Things system can’t effectively identify the data anomaly and locate the anomaly on which layer quickly. A deep recursive network is proposed to model the anomaly detection of the Internet of Things system. The perceptual layer,transmission layer and application layer are used as the output layers of the depth network. The deep recursive network can extract the high-order feature by means of kernel function transformation,and the time sequence characteristic of the deep recursive network itself can improve the accuracy of the anomaly detection. The experimental results show that the deep recursive network can obtain high detection accuracy in anomaly detection of Internet of Things system.