The Research And Design Of Intrusion Detection System Based On PCA And Improved BP Neural Network

The increasing popularization and development of internet brings us great convenience with its rich information resources, and meanwhile, it also brings us network security problems. Conventional network security techniques such as firewall and encryption technique have limited defense effects, and all of them belong to the category of static security techniques whose implementation and maintenance need manual work. As an active security defense technique, intrusion detection offers real-time protection against interior or exterior attack and mistaken operation. It has become a research focus in information security field. However, with the rapid development of intrusion technology, traditional intrusion detection system can not fully meet the need of current network environment, therefore advanced technologies such as neural network and genetic algorithm must be used.This paper opens with some relative conceptions and theories of intrusion detection system. The paper analyzes classical intrusion detection system models and techniques, discovers much of its drawbacks such as high missing report rate. Then study on neural network, the paper finds that neural network can take its advantages such as self-learning and distributed storage when used in intrusion detection system. An intrusion detection system based on neural network will play an important role in the theory and practical if it can be designed and implemented. The paper gives a detailed describing to some improved BP algorithms which will be helpful to fix the drawbacks of original BP algorithm such as low convergence rate and oscillation during learning. According to the high-dimensions of network data, the paper takes principal component analysis to reduce their dimensions before they were sent to neural network inputs. So, the scale of neural network is simplified and computation is reduced. On the base of what was discussed above, the paper put forward a detailed design scheme of intrusion detection model based on neural network, which is made up of five modules. Great emphasis is put in each key module. Lastly, experiment is finished by using the training and testing samples selected from KDD-99 data sets simulate network environment.The experimental results under MATLAB show that: (1) These improved BP algorithms accelerate the convergence rate significantly and enhance the real-time property of the system. (2) The missing report rate of each improved algorithms is very low, which indicates the neural network combined with principal component analysis has a very great advantage in intrusion detection. (3) There are obviously recognition differences between algorithms when processing the same attack data. So, a thought of algorithm optimization is proposed, which opens a door to the future research.