摘要
提出了一种适用于硬件资源受限的传感器节点的坡面异常信号检测方法,数据的采集和异常检测在节点前端完成,能提高监测实时性,降低数据传输的通信能耗。传感器节点采集坡面加速度数据,通过对训练数据进行经验模态分解得到本征模函数的上包络,并对该包络数据进行高斯建模自适应学习确定检测阈值。对监测数据经验模态分解得到上包络数据,运用3α原则与检测阈值进行比较实现异常信号判别。通过算法仿真和实际部署测试得到该方法检测精度为98%,具有一定的实用价值。
As for landslide monitoring,this paper presents an anomaly detection method that is suitable for sensor nodes with limited hardware resources. Data acquisition and anomaly detection can be directly run in the node,which improves the real-time of detection and reduces the communication overhead of data transmission. The sensor nodes collect the acceleration data of the slope,and then obtain the upper envelope of the intrinsic mode function of the training data via empirical mode decomposition. The detection threshold is determined by the adaptive learning based on Gaussian modeling. Similarly,the upper envelope data of the monitoring data can be obtained,and the data will be compared with the detection threshold by 3α principle to perform the anomaly detection. It can be found that the detection accuracy is around 98% through the algorithm simulations and the actual deployment tests,which means that the proposed method is feasible for the practical application.
作者
柳治谱
王勇
LIU Zhipu, WANG Yong(School of Mechanical and Electronic Information, China University of Geosciences, Wuhan 430070, Chin)
出处
《传感技术学报》
CAS
CSCD
北大核心
2017年第10期1536-1541,共6页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金项目(41202232)
关键词
无线传感器网络
滑坡监测
经验模态分解
自适应学习
wireless sensor network
landslide monitoring
empirical mode decomposition
adaptive learning