IoT Hierarchical Intrusion Detection Model Based On Stacked Denoising Autoencoder And Dimension Reduction

With the wide rise of Internet of Things technology in research and application,the issue of network security in the Internet of Things is getting more and more attention.The security problem in the Internet of Things is more serious than the traditional network security problems.It will not only affect the information data,but also cause damage to specific access devices.Internet of Things intrusion detection technology is a very important technical means in the security technology of the Internet of Things.Traditional intrusion detection technology is difficult to meet the heterogeneity of the Internet of Things.In light of this situation,the combination of intelligent learning algorithms and Internet of Things intrusion detection technology has emerged.This paper applies deep learning to IoT intrusion detection,builds a deep network model through noise reduction and self-encoder superposition,and uses unsupervised Greedy layer-by-layer training algorithm to train each layer in order to perform more robustness in the case of noise input.Expression,fine-tuning the model by back propagation.Stacked noise reduction self-encoder can learn more excellent features,so this method can better handle a large number of high-dimensional IoT intrusion detection samples.According to the Internet of Things TCP/IP layered network structure,combined with the characteristics of stacked noise reduction self-encoder,this paper proposes a layered intrusion detection model based on stack-type noise reduction self-encoder to reduce dimension.In the model,a stack-type noise reduction self-encoder(SDAE)is used as a feature of dimension reduction processing,and then the support vector machine(SVM)is used to identify and classify the behavior of the IoT intrusion.Using public data set NSL-KDD to verify the feasibility of the proposed model.In this paper,the performance of the model is compared with other dimensionality reduction models.The results show that SDAE-SVM has better feature learning ability,improves the classification accuracy,and reduces the false positive rate and false negative rate.Comparing the designed IDS of each layer with the full-layer IDS also has higher accuracy,lower false alarms,and lower missed reporting rates,which effectively reduces intrusion detectionload.