Research And Application Of BP Neural Network Optimization Based On The PCA

As more and more Internet comes into people’s lives, internet intrusion becomes increasingly frequent and complicated. The traditional network protection methods can no longer meet the users’needs of Internet safety. Firewall has a lot of shortcomings, although it has been widely seen as an effective method to defend intrusions, but it does not do enough with the increasingly network crisis. So a real-time intrusion detection system is put forwarded to complement Firewall effectively. Only combining various network infrastructures with advanced network security technology can successfully deal with complex mixed-type threats.Aiming at newer network intrusions as they arise, this paper proposes a method that combining misuse detection with anomaly detection, it can be used in intrusion detection system to achieves high accuracy. The anomaly detection using BP neural network module, its own highly self-learning and adaptive capability not only enable the system to recognize intrusion accurately based on existed instances in training samples, but also can recognize and summarize new features of intrusion. Using particle swarm optimization algorithm(PSO) to improve the existing problem of slower convergence speed and local optimal solution of BP neural network. Through global searching ability of PSO to optimize link weight and threshold value to some extent to improving generalization ability and learning performance of BP neural network, so that global search efficiency will be improved. Because intrusion detection data have a characteristic of high dimension in general, this article use PCA algorithms to extract features of input data, so data dimensionality is reduced, the architecture of neural network is simplified, the amount of computation is cut down and the detecting precision is increased. At the same time, the neural network module use methods of handling data packet with different protocols separately.The main direction and structure of the paper is determined by user’s demand for network security and direction of intrusion detection development. This paper first briefly introduces newly proposed system model, which including six modules of data packet model, protocol parse model, pre-process module, rule matching module, PCA-PSO-BP neural network module and response module, then it gave detailed analysis of the structure and shortcoming of BP neural network, and using PSO to optimize BP neural network and abstracting principal components from input data through PCA. Targeted test has been done to TCP and UDP neural network, the results show that it has higher accuracy.