一种提高云计算环境下检测入侵者速度和正确率的新方法

A NEW METHOD FOR IMPROVING THE SPEED AND ACCURACY OF DETECTING INTRUDER IN CLOUD COMPUTING ENVIRONMENT

云计算环境要求入侵检测系统(IDS)极其快速和准确,用于云计算的智能型IDS——反向传播神经元网络(BPNN)经常出现“泛化问题”,即BPNN无解或总误差函数不能收敛于全局最小值。泛化问题降低了BPNN的识别速度和正确率。为了解决该问题,提出两种解决方法。第一种是剔除相关性大的那些特征,保留相关性较小或互相独立的重要特征,以便减少特征数量。第二种是综合利用遗传算法(GA)和最速下降法(梯度法)的优点,降低BPNN的泛化性。用GA求出BPNN全局最优解的近似值,把该近似值作为最速下降法的初始值,进行迭代,最终求得BPNN全局最优解的精确值。仿真实验表明给出的方法是有效的且能够解决BPNN的“泛化问题”,同时能够提高BPNN识别入侵者的速度和正确率。

Cloud computing environment requires intrusion detection system(IDS)to be extremely fast and accurate.Intelligent IDS,back propagation neural network(BPNN),for cloud computing often has generalization problem,that is,BPNN has no solution or the total error function cannot converge to the global minimum.The generalization problem reduces the speed and accuracy of BPNN.To solve the problem,2 solutions are proposed in this paper.The first is to eliminate the features with high correlation and retain the important features which are less relevant or independent of each other,so as to reduce the number of features.The second is to reduce the generalization of BPNN by using the advantages of genetic algorithm(GA)and steepest descent method(gradient method).GA is used to get the approximate value of the global optimal solution of BPNN.The approximate value is taken as the initial value of the steepest descent method,and the exact value of the global optimal solution of BPNN is finally obtained.The experiments show that the method is effective and can solve the generalization problem of BPNN,and it can improve the speed and accuracy of BPNN to recognize intruders.