Detection DDoS of Attacks Based on Federated Learning with Digital Twin Network

Dingling Su, Zehui Qu

Abstract: With Intrusion detection algorithms based on deep learning becoming a hot research topic, most studies pay attention to improving detection accuracy but ignore the problem that can not train a high-precision model due to the limited data of each client. This paper proposes an intrusion detection method based on federated learning and the LSTM model to protect the privacy and improve the classification effect in limited data. As a result of the experiments carried out on the KDD CUP 1999 dataset containing the current DDoS attack types, it was observed that the attacks on network traffic were detected with up to 99.17% success. Furthermore, the federated learning model was constructed in the Digital Twin Network (DTN), an emerging network that utilizes digital twin (DT) technology to create the virtual twins of physical objects. It can real-time monitor the status of physical entities and feedback information to entities in time. Meanwhile, we propose a new optimization framework based on FedProx to tackle the system and statistical heterogeneity inherent in federated networks. This framework shows significantly more stable and accurate convergence behaviour and higher detection accuracy than FedProx and FedAvg.