Research And Application On Abnormal Traffic Detection Algorthm Of Internet Of Things Devices Based On Deep Learning

With the vigorous development of information science and technology,Internet of Things(Io T)devices are widely used in all walks of life,however,known or unknown network attacks bring security risks to Io T devices that cannot be ignored.In order to ensure the security of Io T devices,anomaly detection of their traffic data is required.Among the various anomaly detection methods,the detection methods based on machine learning not only rely on manual collection of attack features,but also need to update the feature library according to new attacks,which is difficult to adapt to the Io T needs of high efficiency and accuracy.In addition,due to the wide variety of Io T devices and the diversification of application scenarios,their activity patterns are significantly different.This difference directly leads to significantly different traffic characteristics,which increases the challenge of feature extraction for anomaly detection methods.Therefore,in different application environments,how to effectively detect abnormal device traffic is of great practical significance for ensuring the security of Io T infrastructure.In order to solve the above problems,ensure the security of Io T devices and satisfy different application scenarios.This paper introduces unsupervised deep learning,and designs two different feature extraction schemes.The specific research contents and results are as follows:(1)An anomaly detection method based on statistical features and attention mechanism is proposed for intelligent Io T devices with large differences in traffic data.Using the statistical feature extraction method,from one-dimensional and two-dimensional perspectives,110 features are extracted from traffic data,so that traffic data information can be more fully expressed.The attention mechanism is further used to give weights to the extracted multi-dimensional features,so as to avoid the model unable to express important information due to too many feature dimensions.Experiments show that the extracted multi-dimensional features are helpful for the model to judge abnormality.Compared with the mainstream models,the proposed method can further improve the abnormality detection effect.(2)Aiming at lightweight Io T devices with strong periodic characteristics,an anomaly detection method named Haar AE is proposed,which introduces wavelet transform to extract the complex periodic feature between the time domain and frequency domain of traffic data,and constructs an anomaly detection method based on Haar AE.The cascade network structure of the convolutional long short-term memory network and the decoder increases the information that the autoencoder can capture the the decoding process and improves the fitting ability of the model.Experiments show that the anomaly detection accuracy of Haar AE in different datasets is higher than that of traditional and advanced anomaly detection algorithms,and it has better robustness.(3)Finally,based on the above two models,this paper constructs an anomaly detection system for Io T device.The system includes a series of modules,such as data storage and visual display.After experiments and system tests,it is verified that the system has certain usability.