Research On Online Anomaly Monitoring Method For IoT Sensor Nodes

With the rapid development of the application of the Internet of Things,the monitoring of anomalous conditions of IoT sensor nodes to ensure system reliability has become a new direction of current research on IoT technology.The anomaly of the IoT sensor nodes includes two reasons: the software and hardware failure of the sensor itself and specific events in the sensor deployment environment.The existing research on anomaly detection of IoT system mainly focuses on the anomaly detection of the sampled data of IoT sensor nodes or the diagnosis of the failure of the node device.Therefore,it is of great practical significance to study methods to detect abnormal states of IoT nodes and identify the source of abnormal states,so that the IoT system can make timely decision-making for node anomalies to improve information credibility.Based on current anomaly detection methods based on pattern recognition,this thesis proposes an online monitoring method for abnormal status of IoT sensor nodes.Combining the spatio-temporal correlation of IoT data,the clustering method and fuzzy logic system are used to realize the online detection of node anomalies and the online identification of node abnormal types respectively.This thesis is supported by the National Key Research and Development Project of China(No.2018YFC0808302).The main contents of this thesis include:(1)Aiming at the problem of abnormal state detection of IoT nodes,a cluster-based method for online detection of IoT node abnormalities was proposed.A composite time series similarity measure criterion is studied.A density-based clustering method based on improved parameter adaptive determination is studied.Using the time correlation of IoT data,the abnormal state detection for a single sensor is achieved through the training phase and the detection phase.Different from the ‘one-time' clustering method for all nodes,the data dimension clustering effect is reduced accurately,and the dynamic performance can meet the requirements of real-time detection of abnormal sensor nodes in practical applications.(2)Aiming at the problem of identifying the source of anomalies in IoT nodes,a method for identifying the sources of anomalies in IoT nodes based on fuzzy logic is proposed.Using the spatial correlation of IoT data,it is confirmed that the abnormal state of the node originates from a fault or event by evaluating the spatial correlation degree of the abnormal node.Extracted the geometric features of node spatial correlation coefficients,researched the fuzzy language set and fuzzy membership function,established a spatio-temporal correlation fuzzy rule base,and finally designed a cascaded fuzzy logic system to calculate the spatial correlation index of abnormal nodes to achieve identification of error and event online.(3)Aiming at the problem of verifying the anomaly detection method of the IoT data set without standards,it is proposed to artificially simulate the abnormal states of the IoT nodes,inject fault nodes and event nodes into the real data set,and generate a detection data set with node anomaly tags.The experimental methods for online detection of anomalous nodes and online identification of anomaly sources are experimental methods to verify the detection capabilities respectively,which provides ideas for algorithm research in the field of anomaly detection in unsupervised learning.For the above work,the actual traffic detection system data is used for experiments and analysis.