无线传感网中一种基于支持向量机的异常事件检测方案

AN SVM-BASED ABNORMAL EVENTS DETECTION SCHEME IN WIRELESS SENSOR NETWORKS

异常事件检测问题是无线传感器网络中的研究热点之一。针对现有检测方案的不足,设计一种新的时间-空间-属性单类超球面支持向量机来建模异常事件检测问题,然后提出无线传感器网络在线和部分在线离群点检测算法。该算法根据节点间的时间-空间和属性关联度确定超球面的半径,最后以在线方式鉴别到达节点的每一个新的测量值是正常数据还是异常数据。仿真实验结果表明,与基于时空关联度的超球面支持向量机相比,新算法的检测率大大上升,虚警率明显下降。同时,部分在线算法与在线算法的效率相当,大大降低了计算和通信复杂度。

Abnormal events detection is one of the research focuses in wireless sensor networks. Aiming at the disadvantages of existing detection schemes, we design a new spatiotemporal-attribute one-class hypersphere SVM （STA-HS-SVM） to model the abnormal events detection problem, and present the online and partial-online outlier detection algorithms for WSNs as well. The algorithms determine the radius of hypersphere according to the spatiotemporal and attribute correlations between the nodes, and finally identify in the way of online whether every new t arriving at the nodes is the normal data or the abnormal data. Simulation experimental results indicate that the new algorithms have significant increase in outlier detection rates and remarkable reduction in false positive rates than the spatiotemporal correlation-based hypersphere SVM. Meanwhile, the partial-online algorithm has similar efficiency as the online algorithm, thereby the computational and communication complexities are significantly decreased.