Research On Malicious Traffic Detection Of Internet Of Things Based On Federated Learning

Under the catalyst of the epidemic,the cloud computing market is exploding,and at the same time,the Internet of Things(Io T)technology has been developing rapidly,and Io T terminal devices are gradually being widely deployed in application scenarios such as smart home and smart agriculture.At the same time,network attacks against Io T devices are increasing,which seriously threaten people’s property and privacy security.And network traffic,as a carrier of behavior between users and devices,is increasingly used by researchers to detect malicious network attacks.With the promulgation of national privacy and security protection regulations and the awakening of people’s awareness of their own privacy information protection in the information age,the demand for protection of data involving user privacy is also increasing.The network traffic data contains a large amount of personal privacy information,and if the data is collected to the server side and trained in a centralized way for malicious traffic detection models,it will involve the risk of violating users’ privacy.How to detect Io T malicious traffic while complying with stricter privacy protection regulations and protecting users’ personal privacy is the primary problem facing the field today.Based on this purpose,this thesis firstly establishes a federated learning based Io T malicious traffic detection algorithm and synthesizes it with other machine learning algorithms,which is the first research of this thesis.Secondly,this thesis explores three problems that would exist if the traditional federated learning algorithm is directly applied to Io T malicious traffic detection,including the heterogeneity of Io T devices,the difficulty of data labeling,and the data drift and concept drift faced during the continuous update of network traffic.Therefore,in this thesis,the following three research works are carried out from the practical problems faced in Io T malicious traffic detection.(1)To address the privacy leakage problem in traditional machine learning-based Io T malicious traffic detection,a federated learning-based Io T malicious traffic detection algorithm is proposed,and a comprehensive analysis based on real Io T datasets is conducted to confirm the effectiveness of federated learning applied to Io T malicious traffic detection.(2)To address the problem that heterogeneity training and unsupervised clients cannot join collaborative training in traditional federated learning,a semi-supervised federated learning algorithm based on knowledge distillation is proposed.The algorithm is used by multiple clients to collaboratively train high-quality malicious traffic detection models,and also solves the problem that clients cannot build heterogeneous models by saving the global model to the client and mutual training with the local model.The results of multiple experiments based on real world Io T network traffic datasets show that the proposed method provides a new research idea for semi-supervised federated learning algorithms as well as semi-supervised malicious traffic detection algorithms.(3)The federated continuous learning problem in network traffic scenarios is defined for the first time,and a personalized continuous federated learning algorithm is proposed for this problem,which effectively exploits the inter-user similarity among clients with non-independent homogeneous distribution by aggregating personalized models for each client with the similarity among client models as the aggregation weight at the server side.The final experiments show that the method enables clients to maximize learning from other clients and alleviates the catastrophic forgetting problem in the continuous learning process.In summary,this thesis proposes an effective solution to the problem of Io T malicious traffic detection based on federated learning,and demonstrates the effectiveness of the proposed method through experiments,which has certain theoretical significance and application value for advancing the development of the field of Io T malicious traffic detection.