云框架超大规模资源处理下无线传感网络数据异常检测

Data Anomaly Detection in Wireless Sensor Networks Under Very Large-Scale Resource Processing in Cloud Framework

为解决无线传感网络数据异常检测方法精度低、网络能量消耗大等问题,提出基于云框架的大规模无线传感网络数据异常检测方法。在云框架超大规模资源处理下,通过自适应过采样法消除冗余数据;运用数据集成、标准化等预处理数据样本,计算协方差矩阵,从高到低排列特征矢量贡献率;采用二进制粒子群方法优化初始森林中的孤立树,挑选初始森林中精度最高、差异度最大的孤立树组建最佳孤立森林,通过异常分数法检测数据特征异常情况,输出对应样本异常概率。实验部分对所提方法、图信号处理法和分层聚合法进行检测率、虚警率、接电能耗、耗时等指标对比,结果表明,所提方法有效降低了节点能耗,数据异常检测速率快、正确率高,具备优秀的适用性与可靠性。

In order to solve the problems of low accuracy and high network energy consumption of wireless sensor network data anomaly detection method,a large-scale wireless sensor network data anomaly detection method based on cloud framework is proposed.Under the super large-scale resource processing of cloud framework,the redundant data is eliminated by using adaptive oversampling method.Preprocessing methods of data samples such as data integration and standardization are used to calculate the covariance matrix and arrange the contribution rate of feature vectors from high to low.The binary particle swarm optimization method is adopted to optimize the isolated tree in the initial forest,the isolated tree with the highest accuracy and the largest difference of the initial forest is selected to form the best isolated forest,the anomaly of data characteristics is detected through the anomaly score method,and the anomaly probability of the corresponding sample is outputed.The detection rate,false alarm rate,power on energy consumption,time-consuming and other indicators of the proposed method are compared with those of graph signal processing method and hierarchical aggregation method.The results data show that the proposed method effectively reduces the node energy consumption,has fast data anomaly detection rate and high detection accuracy,and has excellent applicability and reliability.