基于深度学习的入侵检测研究
|
摘要
为提高网络异常入侵的检测效果,针对传统BP神经网络在检测速度、精度、复杂度等方面的缺陷,提出一种基于深度置信网络的入侵检测方法,该方法通过双层RBM结构降维,再用BP神经网络反向微调结构参数,以达到简化数据复杂度的目的,减少了BP神经网络的计算量。它对传统对比散度训练算法进行了改进,提出了一种改进的对比算法,提高对最优特征的选择结果。仿真实验结果表明,新的网络入侵检测方法较传统的训练算法,能够获得更好的入侵特征提取结果。因此,新的网络入侵检测方法提高了异常入侵检测的检测率,加强了网络安全。
In order to improve the detection effect of network intrusion, this paper proposes an intrusion detection method based on deep belief network for the defects of traditional BP neural network in terms of detection speed, accuracy and complexity. This method reduces the dimension by double-layer RBM structure. The BP neural network is used to inversely fine tune the structural parameters to simplify the data complexity and reduce the computational complexity of the BP neural network. It improves the traditional contrast divergence training algorithm. An improved contrast algorithm is proposed to improve the selection of optimal features. The simulation results show that the new network intrusion detection method can obtain better intrusion feature extraction results than the traditional training algorithm compared with DBN algorithm and BP algorithm. Therefore, the new network intrusion detection method improves the detection rate of abnormal intrusion detection and strengthens network security.
In order to improve the detection effect of network intrusion, this paper proposes an intrusion detection method based on deep belief network for the defects of traditional BP neural network in terms of detection speed, accuracy and complexity. This method reduces the dimension by double-layer RBM structure. The BP neural network is used to inversely fine tune the structural parameters to simplify the data complexity and reduce the computational complexity of the BP neural network. It improves the traditional contrast divergence training algorithm. An improved contrast algorithm is proposed to improve the selection of optimal features. The simulation results show that the new network intrusion detection method can obtain better intrusion feature extraction results than the traditional training algorithm compared with DBN algorithm and BP algorithm. Therefore, the new network intrusion detection method improves the detection rate of abnormal intrusion detection and strengthens network security.
引文
[1]罗守山.入侵检测[M].北京:北京邮电大学出版社,2004.
[2]李春林,黄月江,王宏,等.一种基于深度学习的网络入侵检测方法[J].信息安全与通信保密,2014(10):68-71.
[3]许戈静.基于机器学习方法的入侵检测技术[J].信息通信,2015(12):127-128.
[4]赵海阳,邓京京.基于入侵检测概念、过程分析和布署的探讨[J].内江科技,2007(10).
[5]鄢华.模糊深度学习网络算法的研究[D].哈尔滨:哈尔滨工业大学,2012:14-25.
[6]安琪.基于深度置信网络的入侵检测研究[D].兰州:兰州大学,2016:32-36.
[7]陈达.基于深度学习的推荐系统研究[D].北京:北京邮电大学,2014:17-22.
[8]张春霞,姬楠楠,王冠伟.受限波尔兹曼机[J].工程数学学报,2015,1(2):159-173.
[2]李春林,黄月江,王宏,等.一种基于深度学习的网络入侵检测方法[J].信息安全与通信保密,2014(10):68-71.
[3]许戈静.基于机器学习方法的入侵检测技术[J].信息通信,2015(12):127-128.
[4]赵海阳,邓京京.基于入侵检测概念、过程分析和布署的探讨[J].内江科技,2007(10).
[5]鄢华.模糊深度学习网络算法的研究[D].哈尔滨:哈尔滨工业大学,2012:14-25.
[6]安琪.基于深度置信网络的入侵检测研究[D].兰州:兰州大学,2016:32-36.
[7]陈达.基于深度学习的推荐系统研究[D].北京:北京邮电大学,2014:17-22.
[8]张春霞,姬楠楠,王冠伟.受限波尔兹曼机[J].工程数学学报,2015,1(2):159-173.