采用压缩感知和GM(1,1)的无线传感器网络异常检测方法

An Anomaly Detection Method for Wireless Sensor Networks Based on Compressed Sensing and GM(1,1)

针对当前无线传感器网络(WSNs)异常检测算法的检测准确率较低同时影响网络能耗均衡的问题,提出了一种基于改进压缩感知(CS)重构算法和智能优化GM(1,1)的WSNs异常检测方法。首先,通过建立双层异质WSNs异常检测模型,并采用压缩感知技术对上层观测节点收集到的下层检测节点温度测量数据进行处理,同时结合温度数据稀疏度未知特点,构造有效的稀疏矩阵和测量矩阵,并重新定义测量矩阵正交变换预处理策略,使得CS观测字典满足约束等距(RIP)条件;其次,重新定义了离散蜘蛛编码方式,蜘蛛种群不断协同进化,以获得稀疏结果中非零元素的位置信息,利用最小二乘法得到非零元素的幅度信息,实现了对未知数量检测节点数据的精确重构。在此基础上可以由蜘蛛种群迭代进化得到优化后GM(1,1)的参数序列,通过检测参数序列的相关阈值来判定节点是否发生异常。实验仿真结果表明,与OMP-IGM等异常检测方法相比,该方法的异常检测准确率提高了约7%~33%,网络能耗降低了约18%~43%。

A new anomaly detection scheme for wireless sensor networks （WSNs） based on an improved reconstruction method of compressed sensing （CS） and the intelligent optimizing GM （1, 1） is proposed to improve the accuracy of existing anomaly detection algorithms and to reduce the network energy consumption. A double WSNs heterogeneous anomaly detection model is established, and the CS technology is used to process the upper observation nodes data collected from lower detection nodes. An effective sparse matrix and a measurement matrix are constructed by combining the anomaly detection model characteristics, then an orthogonal transformation pretreatment strategy is redefined for the measurement matrix such that the observation dictionary of CS satisfies the restricted isometry property （RIP）. Since the data sparsity for CS is unknown, a new CS reconstruction algorithm based on discrete social spider optimization algorithm is proposed to realize the accurate reconstruction of the detection node data, and an improved GM （1, 1） intelligent optimization scheme for anomaly detection is designed to achieve a reliable prediction of abnormal nodes in the network. The parameters of GM （1, 1） are optimized through the iteration of the spider population, and abnormalities of nodes are determined by detecting the relevant thresholds of the parameter sequences. Experimental simulation results and comparisons with other anomaly detection algorithms show that the accuracy of the proposed scheme increases by about 7% to 33%, and the network energy consumption reduces by about 18% to 43%.