Study Of Intrusion Detection Based On BP Neural Network And Clustering Analysis

As we all know, computer and Internet are developing rapidly. This gives our life a revolutionary change. At the same time, the network security problems are more and more remarkable. It is a serious threat to the country and people’s life and property safety. Because intrusion detection can block the attack voluntarily and effectively guarantee the network security. Therefore, the study of intrusion detection has important theory value and practical significance.This paper analyzes the disadvantages of BP neural network in intrusion detection, such as the detection of a long time and easy to fall into local minima. At present, although there have been many studies to improve the BP neural network algorithm, basically improve the weights and thresholds for BP network. They improved the detection results. Few people study on BP neural network structure. This paper uses the genetic algorithm not only adjust the weights and threshold of BP neural network, but also optimize the structure of the neural network. The results of MATLAB simulation experiment suggests that the methods has a better accuracy and convergence than BP algorithm and PCA-BP algorithm, and the detection rate and false alarm rate of intrusion system are obviously superior to the traditional methods.This paper introduces the clustering analysis method applied in intrusion detection system. This paper studies the advantages and disadvantages of K-means algorithm in intrusion detection and research status. It finds that some improve strategies only consider the distance factor or density factor when calculate the similarity. On the contrary, this paper not only considers the distance but also the density factor. It makes calculation can find pile in any shapes and reduces the influence of the ac node to latest clustering results. On opinion and confirmation of cluster center, this paper is based on maximum and minimum to improve to select and adjust cluster center. On option and pretreatment parts, using genetic feature selection results of BP neural network as the initial data set, and then reference combining KPCA technology and improved K-means algorithm. And experimental compare, discovered in this paper the improved method of K-means algorithm than the traditional K-means algorithm obviously improve the detection rate and false alarm rate is significantly reduced. It can effectively detect unknown attacks.Generally speaking, research of this paper has more import practical values and theoretical basis for us to analyze intrusion system.