Research On Distributed Data Anomaly Detection In Internet Of Things

The Internet of Things has received a lot of attention over the past few years.Sensors in IoT are vulnerable to malfunctions and malicious attacks,and sensor readings are inaccurate and unreliable,which makes IoT vulnerable to outliers.Outliers are generally considered to be those sensor readings that deviate significantly from normal behavior.The sensor energy in the IoT is limited,and the distributed anomaly detection method can save more energy than the centralized anomaly detection method.Existing anomaly detection methods are either fast but low in accuracy,or high in accuracy but computationally expensive.The general outlier detection methods in the existing methods only use a certain anomaly detection method,which cannot achieve the balance of sensor energy consumption,real-time performance and detection accuracy.Considering multiple contexts,there are some outliers in IoT that are not anomalous events that you want to monitor.For example,in the scenario of forest fire monitoring,the generation of sensor outliers may not be caused by a fire,but by a sensor failure,or by direct sunlight on a certain sensor.Therefore,considering real-time,detection accuracy balance and other issues and multi-context scenarios,this research is carried out.Aiming at the balance between real-time performance and detection accuracy,this paper proposes a distributed anomaly detection based on mixed low-precision and high-precision in the Internet of Things.Through distributed anomaly detection,the information transmission between nodes is reduced to save sensor energy and improve the real-time performance of anomaly detection.Combining multiple low-precision anomaly detections with a small number of high-precision anomaly detections,it is possible to maintain low energy consumption and low time cost with high accuracy;interact with neighbor nodes,and combine neighbor nodes for anomaly detection to distinguish abnormal events and Common noise anomalies.Experiments demonstrate the effectiveness of the proposed method,and the accuracy and F1 score are greatly improved compared to a single conventional anomaly detection method.Aiming at the problem that some outliers are not abnormal events that you want to monitor in multi-context situations,this paper proposes IoT multi-context anomaly detection.This method uses multiple contexts to assist in judging whether an abnormal event has occurred.First,the sensor data at the adjacent moment of the detection data is used to extract features as the moment context,and the unsupervised algorithm we choose can extract corresponding data from different seasons in different seasons.Characteristics of seasons,as seasonal context.And build a community for the sensor,by interacting with the sensors in the same community to detect common outliers,the sensors in the same community have similar attributes,and the sensor data in the same community is used as the spatial context to determine whether an abnormal event has occurred.Experiments demonstrate the effectiveness of the proposed method,the accuracy and F1 score are greatly improved compared to the anomaly detection method that does not utilize multi-context,and it can accurately detect abnormal events.