Integrated Learning ICPS Intrusion Detection Model of Deep Auto-encoder Network

Recent network data under the current industrial cyber-physical system presents more significant and complex characteristics. The traditional use of machine learning to extract features is cumbersome and computationally expensive, which is not conducive for real-time industrial network data traffic detection. Given this, an integrated learning intrusion detection model based on a deep autoencoder network is studied. First, a deep autoencoder network constructed by stacking multiple regular denoising autoencoder networks is used to reduce the data feature dimension non-linearly to obtain a new low-dimensional feature set. Furthermore, multiple deep belief network models are integrated to classify and identify the reduced feature using voting. Compared with traditional dimensionality reduction methods and intrusion detection methods, the experimental results showed that the proposed method performed better in classification and detection applications.