Research About Intrusion Detection Based On Homotopy BP Algorithm

Intrusion detection is an important constant of the network security. Compared with the traditional and still defense way, it is an active and effective network security mean. It can identify and get trace of network attacks, even offer some defense measures. In a result, more and more people pay attention to it. It is believed that the intrusion detection based on neural network algorithm has a bright prospect because of its self-adaptability and self-learning, especially to those complicated data.The article is composed of intrusion and detection. First, it expresses briefly the basic principle of intrusion detection and neural network, and simulates intrusion by NS so appropriately and directly that people can easily understand this process. Later, the article will focus on the detection which includes the advantages and disadvantages of the former BP algorithm analyzed in this paper, and some measures are given to resolve them: First, improve the error function, in order to solve many problems such as stability performance and local minimum point, it is reasonable to replace Mean squared error performance function by cross-entropy. Second, adjust the activation function appropriately and optimize the converging process. Third, in order to accelerate convergence speed, the article relies on homotopy equation and momentum to improve weight. At the same time, find the key point between improved homotopy BP algorithm and intrusion detection, and establish an appropriate MATLAB algorithm platform to test the effectiveness. Comparing the performance of the improved homotopy BP algorithm with the performance of BP with momentum algorithm, it is obvious that the improved homotopy BP algorithm have a faster convergence and lower local minimum points. So it is concluded that the improved homotopy BP algorithm is superior to the BP with momentum algorithm.