Classification Of Abnormal Trajectories Under Collaborative Fog Computing Architecture

In recent years,with the rapid deployment of 5G and the Internet of Things,there are more and more Io T sensors at the edge of wireless networks.These large numbers of sensors can be used for security monitoring and wireless environment perception,and their management and the data they collect are inseparable from fog computing and computational intelligence based on artificial intelligence and machine learning.Distributed fog computing facilitates the use of computational intelligence to process edge data in real time.The work of this thesis considers a fog network application scenario with fog computing,in which a layered fog network can quickly analyze the movement trajectories of surrounding mobile robots or mobile users.The mobile robots in the passing scene try to connect with the fog nodes in the fog network and they may subsequently invade or attack the fog network.Intrusion Detection System(IDS)is undoubtedly a key network security technology in such Io T applications.The main work of this thesis is as follows:1)Summarize the basic principles of network intrusion detection system,and analyze the influence of Rayleigh distribution model and lognormal shadow model on the quality of input data.2)We tried to propose a two-layer cooperative fog network.The first layer is clustered communication Io T devices,and the second layer is from the fog node in each cluster to the access point(AP)of the next-generation Wi Fi system.And this Wi Fi system contains multiple cooperative APs and their edge computing nodes.In this architecture,multiple lightweight fog nodes in close proximity to each other can coordinate with each other and perform distributed perception;then through the AP in the next-generation Wi Fi system,data packets are transmitted to the fog data center,which is also an intrusion detection system Control center node.In this scenario,mobile robots are divided into two types with different characteristics,namely trustworthy and trustless.The trajectory of the latter will adopt the label of the abnormal trajectory.The fog node is responsible for sensing the received signal strength(RSS)from the mobile robot,and sending the received signal to the intrusion detection system through the Wi Fi connection for data fusion and data analysis.Therefore,the intrusion detection system only needs to use limited physical layer information and Media Access Control(MAC)address information for feature extraction,thereby completing machine learning training and realizing the classification of abnormal trajectories.The computer simulation results show that the classification accuracy of abnormal trajectories based on the feedforward network algorithm is 70.6% under the data set containing wireless interference.3)We improve the accuracy of intrusion detection and evaluate the generalization ability of machine learning models by introducing more computationally intelligent models,as well as more sample training sets and new test sets.Computer simulation and unbalanced data analysis show that when the quality of the training set is relatively good,the accuracy performance of the Naive Bayes,the feedforward network and the Long Short Term Memory(LSTM)network increase in turn.