With the rapid development of wireless sensor technology, the application of Wireless Sensor Networks (WSNs) is more and more extensive, and has important military value and broad commercial application prospect. However, due to the limited resources of terminal equipment, wireless communication environment and other reasons, it faces severe security problems. This paper mainly proposes an intrusion detection algorithm based on improved AdaBoost-RBFSVM, and designs an intrusion detection system (IDS) for WSNs denial of service (DoS) attack based on the proposed method. In order to make the RBF-SVM algorithm as the AdaBoost weak classifier, the effect of training is achieved. Using the influence of parameter σ to RBF-SVM and the effect of model training error em on the smoothness of AdaBoost weights, the IABRBFSVM algorithm is proposed. On the other hand, after analyzing the DoS attack, the eigenspace for the attack is proposed, and the corresponding intrusion detection system is designed. Through simulation, the proposed IDS can significantly improve the network performance by detecting and removing malicious nodes in the network, from the perspective of detection rate, packet delivery rate, transmission delay and energy consumption analysis, and has the characteristics of simple structure, short computation time and high detection rate.

Introduction

The open characteristics of WSNs deployment area and the broadcast characteristics of wireless communication cause the network to be vulnerable to various external attacks, which seriously threaten the entire network information security and normal use. The solution to this problem is to develop a wireless sensor networks intrusion detection system to ensure the normal operation of the network. In recent years, with the development of machine learning and deep learning, the AdaBoost algorithm has been successfully applied in intrusion detection. AdaBoost algorithm is proposed by Freund and Schapire1 , Sun X and Yan B use this algorithm to combine multiple classifier cascade structures to implement the intrusion detection application of WSNs2 . Aljawarneh S and Aldwairi combine AdaBoost, random forest and other algorithm3 , using the idea of ensemble learning to get high detection rate in intrusion detection, but it lacks pertinence and complex models, and the generalization is not high. Yu Ren4 uses AdaBoost and Support Vector Machine (SVM) to classify intrusion attacks. Experimental results show that this algorithm is more balanced than SVM alone. SVM was originally proposed by Vapnik and other5 , and Aburomman6 proposed an effective method to optimize the parameters of SVM with Particle Swarm Optimization (PSO) to improve intrusion detection accuracy. Compared with the method of Shams, such as7 , the method has little error and improves the detection speed for the DoS attack. Murugan and other8 put forward the method of deploying the SVM weak classifier to the node by using the wireless sensor networks layer structure and detecting the attack by the joint node. This method has a higher detection rate than the method of deploying IDS to the base station. However, due to the variety of routing protocols in WSN, the algorithm is not universality and detection is not high, and it doesn’t solve the problem of how SVM is adjusted to AdaBoost weak classifier. In order to improve the precision of intrusion detection and adjust the RBF-SVM algorithm as a AdaBoost weak classifier, a new intrusion detection algorithm and an intrusion detection system based on the DoS attack are proposed in this paper, mainly reflected in 2 aspects: adjusting the parameter sigma of RBF-SVM and changing the updating rule of AdaBoost weight, the IABRBFSVM algorithm is proposed. For the DoS attack in wireless sensor networks, the corresponding eigenspace is put forward, and the intrusion detection system with alarm threshold is designed by using the algorithm proposed.