Research On Sensor Node Anomaly Detection And Diagnosis Method Based On Task Execution Trajectory Mining

The application of wireless sensor networks in intelligent traffic systems(ITS)greatly improves the real-time and accuracy of traffic surveillance.However,the complexity of the deployment environment and the limitations of the sensor network lead to various unpredictable failure in the applications of sensor networks.Anomaly detection and diagnosis technology,as one of the key technologies of sensor network fault analysis,can effectively detect and localize anomaly information,which improve the system availability.This paper takes the task of the sensor network application as the research object,anomaly detection and diagnosis technology based on the task execution trajectory is studied deeply.The main work of the paper is presented as follows:(1)Based on the complicated concurrency model of sensor node program which consisting of tasks and asynchronous interrupts,and by analyzing the execution characteristics of the sensor node program,an abnormal detection and diagnosis method based on task and taskassociated functions are proposed.the run-time task execution characteristics and taskassociated functions call characteristics of the perceptual node program are obtained by analyzing the performance of task and task-association functions,and combining the characteristics of tasks such as non-preemption,synchronization,and periodicity.(2)An anomaly detection algorithm based on the task execution path is designed By analyzing the characteristics of the task execution trajectory of the sensor node.The algorithm first uses the specified time window to divide the task execution sequence,and counts the task execution frequency of the application program in each interval to construct the task execution trajectory model;then it proposes an anomaly detection model based on OCSVM classification model and significance test,Then we optimize the parameters of the OCSVM classification model by using the task execution trajectory model during normal execution;finally,the effectiveness of the proposed method was verified through case study.The experimental results show that the OCSVM classification model has a low false positive rate and better detection rate;at the same time,the abnormal task can be detected based on the significance test of abnormal intervals.(3)A statistical model of task-associated functions calling frequency is put forward by analyzing the invocation of abnormal task association function.and an abnormality diagnosis algorithm at function level is designed and implemented to detect the abnormal function based on the model.Firstly,the frequency of task-associated functions is counted when the abnormal task is executed,the frequency of task-associated functions of the task execution constitutes the statistical model of task-associated functions calling frequency;using the significance test to compare the function calls of the statistical model of task-associated functions calling frequency during normal execution and abnormal execution,we can determine the abnormal function called by the task.Finally,the fault code is found by analyzing the related call of the abnormal function and the related source code control flow information.The result shows that the algorithm can achieve anomalies positioning and diagnosis.The anomaly detection and diagnosis algorithm based on task execution trajectory mining proposed in this paper firstly performs abnormal task detection based on the task execution trajectory,and further mines the related abnormal information of abnormal task-associated functions calling frequency to detect the abnormal function;then according to the abnormal function information such as call relations,control flow,and data flow enables tracking and positioning of abnormal information.The detection algorithm based on task execution trajectory and the abnormality diagnosis algorithm based on the statistical model of taskassociated functions calling frequency proposed in the thesis narrow the scope of abnormal and provide help for repairing code defects of the sensor node program.