PCA-PSO-BP Neural Network Application in IDS

Abstract

BP neural network has two disadvantages, one is to fall into local minimum value easily; the other is the slow convergence. We propose in this paper an approach, including three main operations. Firstly, the algorithm of particle swarm optimization (PSO) is applied to improve back propagation (BP) neural network. Secondly, principal components analysis (PCA) method is used to deal with the original information. Thirdly, after optimization of BP neural network, we employ it into the intrusion detection system. The simulation results reveal that the new proposed BP neural network is superior to the traditional BP neural network. Introduction Neural network plays an important role in the domain of intelligent control, and it has made great progress in every domain, such as neurosciences, mathematics, statistics, computer science etc. Now, BP neural network application is one of the most widespread applications, this is due to the fact that BP neural network algorithm is simple and plastic. But there exits two disadvantages, falling into local minimum easily and having a slow convergence. For these problems many researchs scholars proposed a lot of solutions that including Conjugate gradient, Newton, Causs-Newton, Levenberg-Marguard methods etc [1] [2] [3] [4]. Also the problems above have a great improved. But the calculated quantity of these methods is relatively large and the methods fail at dealing with large-scale data. In addition, some experts so far have proposed some hybrid intelligent algorithms, like combining BP neural network with artificial immune or particle swarm algorithms. These combinations make BP neural network a better efficiency and more widely application. According to this idea, this paper adopts particle swarm optimization (PSO) to optimize the BP neural network. And then apply it to the intrusion detection. Most of the data from network is multidimensional and noisy. So in this paper, data is processed by principal components analysis method first. The paper is organized as follows. Section 2 introduces the principle of BP neural network and PSO. The detailed design of PCA-PSO-BP neural network is designed in section 3. Intrusion detection system and simulation results are designed in section 4 and section 5. Finally we concludes this paper. BP neural network and PSO introduction Since Rulmhart and parallel distributed processing (PDP) group put forward BP algorithm in 1986[5], and then the artificial neural network researchers began to pay attention on BP neural network. But the biggest flaw of BP neural network is slow learning efficiency and slow convergence. Particle swarm optimization algorithm is put forward by Kennedy and Eberhar in the United States in 1995, a kind evolutionary algorithm based on intelligence [6]. And its aims at to simulate the unpredictable movement of birds. The main idea is to regard a solution from the problem as a particle i . The particle i refers to one bird. The particle’s process of searching for its optimal solution refers to the process of searching for food. The above process can be described in Figure.1. International Power, Electronics and Materials Engineering Conference (IPEMEC 2015) © 2015. The authors Published by Atlantis Press 145 Impact of the current speed Group influence particle self-memory influence ( ) gd p t ( ) 1 id v t + ( ) 1 id x t +