CAJ | 학술논문

为克服BP算法易陷入局部最小的缺点,同时为减少样本数据维数,提出一种基于主成分分析(PCA)的遗传神经网络方法.通过降维和去相关加快收敛速度,采用改进的遗传算法优化神经网络权值,利用自适应学习速率动量梯度下降算法对神经网络进行训练.MATLAB仿真实验结果表明,该方法在准确性和收敛性方面都优于BP算法,应用于入侵检测系统中的检测率和误报率明显优于传统方法.
위극복BP산법역함입국부최소적결점,동시위감소양본수거유수,제출일충기우주성분분석(PCA)적유전신경망락방법.통과강유화거상관가쾌수렴속도,채용개진적유전산법우화신경망락권치,이용자괄응학습속솔동량제도하강산법대신경망락진행훈련.MATLAB방진실험결과표명,해방법재준학성화수렴성방면도우우BP산법,응용우입침검측계통중적검측솔화오보솔명현우우전통방법.
In order to reduce the high-dimensions of sample datum and to overcome the disadvantage of BP algorithm which was easy to get into the local least value, this paper presented a genetic algorithm neural network method which based on principal component analysis (PCA). Through reducing dimensions and decorrelation to increase the convergence speed, adopted the improved genetic algorithm to optimize neural network weights, used adaptive learning rate momentum gradient descent algorithm to train neural networks. The results of MATLAB simulation experiment shows that the method has a better accuracy and convergence than BP algorithm, and the detection rate and false alarm rate of intrusion detection system are obviously superior to the traditional methods.