改进差分进化算法优化BP神经网络用于入侵检测  

IMPROVED DIFFERENTIAL EVOLUTION ALGORITHM TO OPTIMIZE BP NEURAL NETWORK FOR INTRUSION DETECTION

作　　者：陈超[1] 曹晓梅[1] Chen Chao;Cao Xiaomei(School of Computer and Software,Nanjing University of Posts and Telecommunications,Nanjing 210023,Jiangsu,China)

机构地区：[1]南京邮电大学计算机与软件学院,江苏南京210023

出　　处：《计算机应用与软件》2018年第4期310-316,324,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61202353);国家重点基础研究发展计划项目(2011CB302903);江苏高校优势学科建设工程项目(yx002001)

摘　　要：为解决BP神经网络应用于入侵检测时检测率较低、训练时间过长的问题,对改进差分进化算法(SAMDE)优化BP神经网络并用于入侵检测的可行性进行研究。该算法引入模拟退火算法(SA)和一种融合DE/rand/1与DE/best/1的变异算子对差分进化算法进行改进以提高其全局寻优能力。用改进后的算法优化BP神经网络权值阈值。通过逐次的迭代训练使BP神经网络收敛,将优化过的BP神经网络用于入侵检测。仿真实验结果显示,优化的BP网络在收敛速度和精度方面有明显提升,用于入侵检测时提高了检测准确率,缩短了训练时间。In order to solve the problem of low detection rate and long training time when BP neural network is used in intrusion detection,the feasibility of improving BP neural network by using improved differential evolution algorithm(SAMDE)and intrusion detection was studied.This algorithm introduced a simulated annealing algorithm(SA)and a mutation operator that combined DE/rand/1 and DE/best/1 to modify the differential evolution algorithm to improve its global search ability.The improved algorithm was used to optimize the weight threshold of BP neural network.The BP neural network was converged by successive iterative training,and the optimized BP neural network was used in intrusion detection.Simulation results show that the optimized BP network has obvious improvement in convergence speed and accuracy.When used in intrusion detection,it improves the detection accuracy and shortened the training time.

关 键 词：BP神经网络 差分进化算法 模拟退火算法 全局寻优 入侵检测 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]