融合DBN和BiLSTM的工业互联网入侵检测方法

针对当前工业互联网的攻击行为复杂,其网络数据具有海量、高维、时序性和非线性等特征,导致传统入侵检测方法的特征提取困难、检测率低、泛化能力差等问题,提出一种融合深度信念网络(deep belief network,DBN)和双向长短时记忆网络(Bi-directional long short-term memory,B...

Industrial internet intrusion detection method integrating DBN and BiLSTM

With the complex attack behavior of the current industrial Internet, such massive, high-dimensional, sequential and non-linear characteristics of its network data lead to the difficulty of feature extraction, low detection rate, and poor generalization ability of traditional intrusion detection methods. In view of these problems, a deep hybrid intrusion detection model is proposed. This model combines deep belief network (DBN) and Bi-directional long short-term memory (BiLSTM) networks. Firstly, the data set is preprocessed; secondly, DBN and BiLSTM are used to extract features of nonlinear features and long-distance dependence information respectively; finally, the softmax classifier is used to identify intrusions. The test results show that compared with the current leading algorithm, the accuracy of this method on the gas pipeline dataset is increased by 0.82%, and the false alarm rate is reduced by 0.35%; the accuracy on the UNSW-NB15 dataset is increased by 2.40%, the false alarm rate is reduced by 4.17%.