The Research Of Intrusion Detectiontechnology In Mobile Ad-hoc Networks Based On Deep Belief Network

Mobile ad hoc networks(MANETs) have been widely used in wireless communication field, but its characteristics make it more vulnerable to all kinds of intrusion. So, it is very valuable to research the security of MANETs. As an active security protection measure, intrusion detection technology is the key to ensuring the security of MANETs. Deep learning was emerged as a new machine learning technology in recent years, mainly explored modeling and learning problems of multi-layer artificial neural network, had got great success in speech and image recognition field, provided a new way to solve such a complex behavior recognition problem of intrusion detection in MANETs.Aiming at the diversity and complexity of security issues in MANETs, this thesis presents a method for intrusion detection based on deep belief network(DBN). As a kind of mature deep learning model, DBN could achieve better detection accuracy, if it is applied to intrusion detection technology in MANETs. The main work and contributions of this thesis were listed as follows:Firstly, combined with the characteristics and security threats of MANETs, this thesis analyzed problems we would face when applied intrusion detection technology in MANETs, and a few typical intrusion detection algorithms. Then we studied the model structure and learning theory of DBN, training algorithm of Restricted Boltzmann Machine, discussed the feasibility of applying DBN to intrusion detection technology in MANETs.Secondly, this thesis designed the structure of intrusion detection system in MANETs based on DBN. Detailed analysis, such as wireless packets capturing, network feature extracting, DBN training and intrusion detection were included. Then we provided solutions of some problems we might encounter in training DBN.At last, the intrusion detection model based on DBN for MANETs was simulated. Aiming at routing attack in MANETs, we realized adding black hole and selfish nodes in NS2. After that, normal MANETs and abnormal MANETs with malicious nodes were simulated, performance of the three kinds of network were analyzed. We extracted the behavior characteristics generated in the process of network simulation. Then, we accomplished the simulation of the intrusion detection model based on DBN in MATLAB. The feature set which was extracted before was used to train and test DBN. Then we analyzed the simulation results, which verified the feasibility and availability of the intrusion detection method this thesis proposed. Compared with traditional BP intrusion detection model, DBN model can achieve better intrusion detection performance.