Research And Improvement On Intrusion Detection Algorithms Based On Deep Learning

With the rapid development of Internet and its wide application,network security becomes an attractive topic for the public.As a crucial technique in information security,intrusion detection adapts the ever-changing cyber-attacks and secures the networks and systems via the active defence.In recent years,deep learning is the focus in academic and obtains the huge achievements in image processing,speech recognition,text generation and so on.Its powerful ability in classification enlightens the update and development of intrusion detection.Increasingly,deep learning algorithms are drawn into intrusion detection.This paper mainly discusses the intrusion detection technologies based on deep learning and designs models from defence and attack.In the aspect of defence,starting from the intrusion detection systems based on traditional machine learning algorithms,the latest deep learning algorithm improves the intrusion detection system.In the aspect of attack,an approach to generating adversarial traffic is proposed,providing a new thought for the optimization of intrusion detection in the future.The main work and innovations are listed as below:First,the intrusion detection models are implemented based on traditional machine learning algorithms.According to metrics in detection,the detection performance is analyzed,supplying reference for the latter researches.Afterwards,the intrusion detection system model based on character-level convolutional neural networks is established depending on deep learning.The model optimizes the preprocessing to data and strengthens the classification capacity for the model.Compared with the model based on traditional machine learning algorithms and common deep learning algorithms,the performance of the model is improved remarkably.Finally,based on the generative model,we propose the malicious traffic generation model based on generative adversarial networks for the first time.Inspired by Game Theory,the adversarial malicious traffic is generated,which is able to evade the detection of intrusion detection systems,used to detect the robustness of intrusion detection.