Research And Application Of Optimized Neural Network In Intrusion Detection

With the inconceivably rapid development of network technology in these years, it brings more convenience to our lives. Changes that brought with the diversified development of Internet are reflected in the people’s life, learning and work. Especially some emerging technologies such as mobile Internet, cloud computing and Internet of Things have brought enormous changes of people’s lives in recent years. But with the development of the network, it often has a negative impact. Particularly the risks of network security make people cannot use the network safely, it seriously affect the development of Internet, lives of people and security of society. Therefore the importance of network security is noticeable.Traditional network security technologies mainly provide static protection and can’t provide protection against internal attacks. The IDS is behind a firewall, detecting the network without affecting network performance, providing protection against internal attacks, external attacks and misuse in real time.In this paper, we researched the application of optimized neural network in intrusion detection deeply. First we proposed a classification method of dataset feature in accordance with the intrusion of ways and compared the detection accuracy of various feature extraction methods, determined the method of feature extraction of this paper. Then we analyzed the BP neural network which was optimized by PSO, improved its shortcomings and presented an improved BP neural network. Finally, we designed an intrusion detection model on based of the feature extraction method and improved algorithm, also verified it does have a better detection accuracy. The main contributions of this paper are as follows:1. Design processing module of intrusion detection dataset. There are two steps of processing the dataset:Data normalization and dataset feature extraction. Data normalization transforms non-digital features into digital ones of dataset feature, because the range of digital features is too large, it needs to reduce the differences. All digital features are mapped to [0,1] according to the formula. For dataset feature extraction, we compared four feature extraction results and selected the highest detection accuracy of feature extraction methods for different intrusions, it was prepared for the follow-up experiments.2. Improve traditional BP neural network which was optimized by PSO. PSO algorithm in terms of global search can cover the shortage that is BP network easy to fall into local optimum value to a certain extent. However, PSO algorithm is still easy to fall into local optimum value in iterative process lead to prematurity of algorithm. Therefore, we presented a method that supervising global optimum particle. The optimal particle subgroups update and reselect global optimum particle, when particle doesn’t update for a long time.3. Present an intrusion detection model. The model based on the feature extraction method and improved BP neural network which was optimized by PSO, It trains network according to four types of intrusion, centralizes the difference between normal data and abnormal data. It improves the overall detection accuracy by improving each neural network. Experiments show that the detection accuracy rate of improved model reaches99.26%. The detection accuracy rate and false negative rate of improved model are obviously better than traditional model. But the improved model also has defects that result in a slightly higher false positive rate.