WSN中基于超椭圆判决边界的异常检测的动态建模

Dynamic Modeling of Anomaly Detection Based on Superellipse Decision Boundary in WSN

为了实现无线传感器网络的动态数据流环境中的异常检测,提出了一种称之为动态数据捕获异常检测的迭代超椭圆判决边界方法;具体实现是建立起异常检测超椭圆模型,每个节点基于到当前时间为止的测量值来调整其超椭圆模型;当边界参数变化较小时,动态数据捕获异常检测算法终止,最终收敛到覆盖正常和异常测量值的超椭圆边界;为了提高模型对监测环境中数据变化的跟踪能力,提出了一种采用遗忘因子并结合滑动窗口的基准估计和有效N跟踪的方法来提高模型在非平稳环境中的跟踪能力,从而实现对数据真实流属性的捕捉;仿真实验结果表明,提出的动态建模方法相比于目前先进的静态建模方法,不仅具有更高的准确性和异常检测能力,而且具有更强的数据变化的跟踪能力和检测能力。

In order to realize anomaly detection in dynamic data flow environment of wireless sensor networks,an iterative hyperelliptic boundary decision method called as dynamic data capture anomaly detection is proposed.The concrete implementation is to establish the anomaly detection hyperellipse model,each node adjusts its hyperellipse model based on the measured values up to the current time.When the boundary parameter changes are small,the dynamic data capture anomaly detection algorithm terminates,and finally converges to the hyperellipse boundary covering the normal and abnormal measurements.In order to improve the tracking ability of the model to the data changes in the monitoring environment,a forgetting factor combined with the benchmark estimation of sliding window and effective N tracking method is proposed to improve the tracking ability of the model in non-stationary environment,so as to capture the real data flow attributes.The simulation results show that,compared with the advanced static modeling methods,the proposed dynamic modeling method not only has higher accuracy and anomaly detection ability,but also has stronger tracking and detection ability of data changes.