Research On Intrusion Detection Technology For Wireless Sensor Networks Based On Unsupervised Methods

Wireless sensor network is a self-organization distributed network system made up of a group of tiny sensor nodes lack of computation, storage and energy through wireless medium. It is widely applied in a great many aspects such as military affairs, environment monitoring, industry, medical treatment, transport and civilian affairs. However, due to limited capacity of energy, computing and storage, dynamic network topology, and large number of nodes easier to invalidate, it suffers intrusion much more easily than the traditional network. It doesn't have the infrastructure, so the firewall technology in the wired network has been unfit to be applied in it. At present, it is difficult to design a better intrusion detection scheme to enhance the security defensive performance for wireless sensor networks combined with safety, energy saving, accuracy and expeditiousness.Intrusion detection technology is an extremely important security tool in computer networks and it has already become a research focus as a second line of security defense to make up for the shortage of intrusion prevention technology. Presently, it has infiltrated into the wireless network environment; whether it is mobile Ad Hoc networks, wireless sensor networks or wireless mesh network all regard IDS as a hot issue to research. There have already been many research results on IDS in the cable network and mobile Ad Hoc networks, however, there is relatively less related study resting on the experimental stage in wireless sensor networks. Therefore, it is quite necessary to do research on IDS in wireless sensor networks, for example, some definite intrusion detection methods and the corresponding intrusion detection model.Nowadays, the wireless intrusion detection system is lack of effectiveness, adaptability and scalability, so it is necessary to construct an intrusion detection model in a more systematic and intelligent approach. Intrusion detection is a problem of pattern recognition and artificial neural network has the ability of self-organization, self-learning and generalizing as an important method of pattern recognition. It enables the system to have both ability of better recognition of known attacks and detection of unknown attacks to apply the artificial neural network in the wireless IDS. Therefore, in recent years, research on intrusion detection based on neural networks has become a focus and will have a bright prospect. The IDS based on neural networks can be divided into four categories: (1) IDS based on the multi-layer feed forward neural network; (2) IDS based on the feedback neural network; (3) IDS based on the Unsupervised neural network; (4) IDS based on the hybrid neural network. This paper just applied the intrusion detection idea based on Unsupervised methods in the field of wireless network, carrying through the study of intrusion detection models and methods.This paper analyzed the methods and models of IDS, combining with the development status quo of IDS in wireless sensor network, and then it put forward an intrusion detection system based on Unsupervised neural network (UNNIDS) aiming at the security characteristics of wireless sensor networks, introducing neural network technology. The work of paper is mainly in the following aspects:(1) It pointed out the necessity and urgency of research on intrusion detection technology in wireless sensor networks, through comparative analysis of their own characteristics of IDS both in wired and wireless environment;(2) Through research on intrusion detection methods in wireless mobile Ad Hoc network, it pointed out that the existing mature intrusion detection techniques of Ad Hoc aren't entirely applicable to wireless sensor networks but deserve reference. IDS in wireless sensor networks could combine feature-based detection and anomaly detection, which will help to provide much more complete IDS in function;(3) It analyzed the superiority and status quo of application of neural networks to intrusion detection field, and at the same time, it indicated the problems during the application;(4) It studied on the existing Unsupervised intrusion detection algorithms, and introduced them in the field of wireless intrusion detection;(5) According to the security requirement characteristics, it put forward an intrusion detection method based on Unsupervised techniques for wireless sensor networks. The method has the following characteristics: 1) it can detect both known and unknown attacks; 2) it can learn input data without prior knowledge; 3) it can adjust self-adaptively according to changes in the environment;(6) After analyzing the characteristics of ART2 neural network, it was applied to Unsupervised neural network engine for UNNIDS model, and simulation experiment was also carried out using KDD CUP'99. The experimental results show that the new method has a higher detection rate, a lower rate of false positives and excellent dynamic, which fully explains the feasibility and effectiveness of the method about unknown intrusion detection.Lots of research on IDS based on Unsupervised neural network has been done, but only limited to cable network. The paper applied the intrusion detection idea based on Unsupervised methods into the field of wireless networks, which is the innovative point. After a large amount of study on both international and domestic information, we hardly find any research on the application of ART2 technique, which is able to preferably solve two difficult problems of "stability and plasticity", in wireless sensor networks. Thus, it is also the innovative point that the paper put forward an intrusion detection engine based on ART2 for wireless sensor networks.