Research And Implementation Of Data Anomaly Detection Method In Sensor Networks

In recent years,along with the rapid development of micro-computer technology,electronic technology and modern network and wireless communication technology,wireless sensor networks have emerged.They have gradually become an important channel for human beings to obtain data and information from the physical world.They have attracted great attention of governments,academia and industry,and researchers pay more and more attention to the security threats and security issues faced.Wireless sensor networks have the characteristics of limited node resources and vulnerability to attack.Although the security of sensor networks has been improved by cryptography,secure routing and other technologies,there is still a lack of effective methods to detect data anomalies in sensor networks,so it is difficult to guarantee the reliability of data in sensor networks.The improvement of anomaly detection technology can greatly promote the future application and development of sensor networks.How to use data efficiently and accurately for anomaly detection is a hotspot in the field of sensor network security.In this paper,the research work on anomaly detection in wireless sensor networks is carried out.We analyses the research results in the field of anomaly detection in sensor networks at home and abroad,then classifies,compares and studies the existing detection technologies.Finally,a hierarchical anomaly detection algorithm based on information entropy and K-means is proposed.Based on this,an anomaly detection system for sensor networks is designed and implemented.The main work of this paper is as follows:1.Aiming at the core problem of sensor data anomaly detection,this paper proposes a single sensor anomaly detection method based on information entropy to calculate the probability of anomaly.The method calculates the information entropy sequence by using the data values in the sliding window of a single sensor as discrete random variables,and then calculates the abnormal probability of the data stream to detect the abnormality of a single sensor combined with the change of the data values of the data stream.This method can integrate the time correlation and statistical probability characteristics of single sensor data stream,compared with the method based on data distance,it can distinguish the normal and abnormal changes of data more accurately.2.Based on the detection method which use information entropy,combined with the existing anomaly detection method based on K-means,this paper proposes a hierarchical anomaly detection algorithm which integrates the detection technology based on information entropy and K-means clustering detection technology.Firstly the algorithm detects the sensor's suspicious anomaly by using the method of outlier probability discrimination,and then confirms it by combining the results of K-means clustering detection algorithm after the sensor's suspicious anomaly is detected by the method of prior detection.Among them,information entropy is not only used to calculate the probability of sensor anomalies,but also to participate in K-means clustering detection as an important feature.The hierarchical algorithm can make more reasonable use of equipment resources,give full play to the performance advantages of equipment,and achieve the balance between detection accuracy and efficiency.Moreover,the algorithm can be flexibly applied to sensor networks through hierarchical collaborative anomaly detection model.3.According to the proposed hierarchical anomaly detection algorithm based on information entropy and K-means,a sensor network anomaly detection system is designed and implemented.The module division and internal design of the system are introduced in detail.The usability of the detection system is tested by simulating anomaly data.