Research On Maritime Network Intrusion Detection Based On Deep Learning And Transfer Learning

The intrusion detection system(IDS)plays an essential role in protecting ICT(Information and communication technology)infrastructure.In the maritime wireless communication network,the nodes are widely distributed,the communication interference between nodes is serious,and the number of nodes changes from time to time,which leads to large data flow fluctuations and changes of network topology,resulting in real-time change of the maritime wireless network environment.In addition,in maritime wireless network,the amount of communication data between nodes is insufficient for the model to be trained.Based on the characteristics above,it is difficult for traditional intrusion detection technology to obtain the same performance in maritime wireless network as in ordinary network.To solve these problems,this thesis focuses on the research of the combination of deep learning algorithm and intrusion detection technology,and focuses on the design of an IDS with high detection rate,wide coverage of network attack types and excellent adaptive ability.The research focuses on the adaptability of IDS to the environment change of maritime wireless network and deal with the lack of training data in maritime wireless network.This thesis proposes corresponding new algorithms.The specific contents are as follows:(1)In order to make it possible for the IDS to adapt to changeable maritime wireless network environment and maintain reasonable detection rate at the same time,we propose an unsupervised self-adaptive deep learning IDS.The sparse coding is used for self-taught learning and the IDS can grasp an attack's nature from the unknown environment and its unlabelled data.Meanwhile,through combining ELM with DBN and using rectified linear unit as the activation function,the algorithm improves the traditional learning framework,which effectively improves the efficiency of model training.The experimental results shows that the adaptability of this method is good enough and it can maintain a high detection rate in the dynamic maritime wireless network environment.(2)The updating of detection model based on deep learning needs a large amount of labelled training data and the collection of labelled data is quite time consuming.Particularly,in maritime wireless network communication,the traffic between nodes is low and the labelled data is difficult to be obtained.In this thesis,we study the feasibility of combining Federated Learning and Transfer Learning,which is a method that can transfer the learned features and knowledge from the source model to the target model with less data requirement.In addition,in order to make full use of the computing power of the nodes in the network and obtain better performance on different nodes,5% of the data obtained by each node is shared for distributed training.We compare the performance of hybrid IDS model based on Federated learning and Transfer learning with traditional IDS model.At the same time,we also compare the performance of IDS model trained by Federated learning with and without data sharing.Experimental results shows that the proposed IDS model performs better in identifying attacks when the available training data is relatively small.Compared with not sharing data between nodes,sharing 5% training data can greatly improve the overall detection rate of the model,and effectively deal with the problem caused by non independent and identically distributed data.