| 摘 要: 近年来,网络安全检测已经取得了很大的进步。然而,网络迅速的发展、流量分布的变化和数据样 本中的噪声等问题都对现有方法提出了很大的挑战。针对此,提出了基于非对称分解卷积的网络安全检测(Network Security Detection Based on Asymmetric Decomposed Convolution,ADC-NSD)方法。ADC-NSD方法根据对网 络连接数据的训练与学习,生成区别常态与危险状态的安全检测模型,通过对卷积神经网络中的卷积核进行分解,完 成对数据进行解析和检测。最后,以KDDCUP99为测试数据集,将ADC-NSD方法与其他机器学习方法进行比较。实 验结果表明,ADC-NSD方法能有效地解决网络安全检测问题,总体精确率为98.72%,准确率为99.92%,召回率为 94.61%,F1值为97.19%。 |
| 关键词: 网络安全;安全检测;卷积神经网络;非对称分解卷积 |
|
中图分类号: TP18
文献标识码: A
|
|
| Network Security Detection based on Asymmetric Decomposed Convolution |
|
FENG Renjun, WU Ji, WANG Zhenyu, JING Dongsheng
|
( Suzhou Power Supply Branch, State Grid Jiangsu Electric Power Limited Company, Suzhou 215004, China)
frj1989@126.com; 13862159678@163.com; gcxy6@hotmail.com; jds19810119@163.com
|
| Abstract: In recent years, network security detection has made great progress. However, the rapid development of communication networks, changes of the traf c distribution and the noise in data samples all pose great challenges to the existing network security detection methods. To solve this problem, an approach referred as Network Security Detection based on Asymmetric Decomposition Convolution (ADC-NSD) is proposed. ADC-NSD generates a security detection model to distinguish normal state and dangerous state according to the training and learning of network connection data, and then analyzes and checks the data through decomposing the convolution kernel of the convolution neural network (CNN). Finally, using KDDCUP99 (KDD: Knowledge Discovery and Data Mining) dataset as testing dataset, ADC-NSD is measured against other machine learning algorithms. The results show that ADC-NSD could be well applied to network security detection. The overall accuracy rating is 98.72%, precision rate being 99.92%, recall rate being 94.61% and F1 score being 97.19%. |
| Keywords: network security; security detection; convolution neural network; asymmetric decomposed convolution |
