基于特征补全的无线传感器网络异常数据流检测

Anomaly Data Flow Detection in Wireless Sensor Networks Based on Feature Completion

由于节点所感知数据有缺失或者错误的情况,使异常数据流检测受困,导致检测准确率、漏报率和能耗等方面存在问题,因此,提出基于特征补全的无线传感器网络异常数据流检测方法。根据传感网络内数据流间的相关性,在特定环境内对缺失和错误数据进行估计与补全;从补全后的无线传感器网络数据流中抽取数据,并完成数据特征挖掘,为之后的异常数据流检测做好准备;使用支持向量机将正常数据和异常数据分隔,从而实现对无线传感器网络异常数据流检测。结果表明:特征补全后的无线传感器网络异常数据检测,其检测的准确率维持在99%以上,漏报率在0.3%以下,能耗下降率最高可达到35.87%,检测用时在0.8 s以下,具有准确率高、漏报率低、能耗少且用时短的优势。

Due to missing or incorrect data perceived by nodes,abnormal data flow detection is hindered,resulting in issues with detec-tion accuracy,false positive rate,and energy consumption.Therefore,a feature completion based abnormal data flow detection method for wireless sensor networks is proposed.Missing and erroneous data in a specific environment are Estimated and completed based on the correlation between data flows within the sensor network.Date are extracted from the completed wireless sensor network data stream and data feature mining is completed to prepare for subsequent abnormal data flow detection.Support vector machines is used to separate normal and abnormal data,thereby achieving abnormal data flow detection in wireless sensor networks.The results show that after fea-ture completion,the accuracy of abnormal data detection in wireless sensor networks is maintained at over 99%,with a false alarm rate below 0.3%,and a maximum energy consumption reduction rate of 35.87%.The detection time is below 0.8 seconds.The proposed method has the advantages of high accuracy,low false alarm rate,low energy consumption,and short time using.