Research On Distributed Deployment Of Anomaly Detection Scheme Based On Internet Of Things

With the development of Internet of things(IoT),more and more devices can realize network interconnection.At the same time,with the increase of the number of devices,the network is facing more and more threats.The resources of the underlying IoT devices are relatively limited,and complex anomaly detection mechanisms cannot be deployed on these devices.At present,most researches deploy anomaly detection mechanisms to central nodes such as servers and edge network management nodes.For the Internet of things whose underlying devices can communicate with each other,the anomaly detection mechanisms deployed in the central node may not be able to detect the communication content between devices.At present,the researches on distributed deployment mostly focus on how to lighten the anomaly detection mechanisms,which may lead to the decrease of detection accuracy.In order to solve this problem,this paper mainly proposes a method of distributed anomaly detection mechanism deployment,which divides a complete anomaly detection mechanism into multiple modules and deploys it in the underlying nodes of the Internet of things.This deployment preserves the integrity of the mechanism and can ensure a better detection effect.At the same time,by dividing the mechanism into deployments,the computing resources needed can be reduced.In this paper,we first study the ideal network situation,aiming at minimizing the cost.We build an optimization model,and use the improved branch and bound method to solve the optimization problem.At the same time,we simulate the algorithm.After deployment,the detection effect of anomaly detection function and the loss of computing resources are analyzed.The simulation results show that,in the ideal network situation,this deployment mechanism can reduce the amount of computing resources needed by nodes under the condition while ensuring better detection effect.Then,this paper studies the special network conditions with more limited resources,gives the changes of the optimization model,and uses the improved genetic algorithm to solve the problem.This paper analyzes the detection effect of anomaly detection function and the loss of computing resources after deployment.The research results show that the anomaly detection deployment mechanism proposed by us can reduce the amount of computing resources needed by nodes while ensuring better detection effect.Finally,this paper studies the mechanism of granularity adjustment of anomaly detection mechanism.This paper presents a dynamic adjustment mechanism of anomaly detection granularity based on reinforcement learning.The mechanism mainly predicts the next network state according to the current network state,and adjusts the anomaly detection granularity accordingly.Different from the existing research,the adjustment mechanism proposed in this paper adopts reinforcement learning algorithm,which can form the model and predict the network state without prior knowledge.At the same time,we simulate the effect of the mechanism,the accuracy of its prediction and the use of computing resources.The simulation results show that this mechanism can reduce the amount of computing resources used by nodes under the condition of good detection effect.