Research On Network Traffic Anomaly Detection Based On Deep Reinforcement Learning

With the rapid progress of Internet and information technology,wired network and wireless network have become an indispensable part of our life.Similarly,the level of network information technology is constantly improving,and network security needs to be paid enough attention to,because this development will lead to some illegal intruders to threaten the network security.For example,they will steal your information,will spread Trojan horse virus to the network and so on.So we should be vigilant to identify these intruders,so as to protect our network security from infringement.With the rapid development of computer technology,network traffic anomaly detection technology has become a variety of technology,these technical means become very intelligent,and also very complex.Traditional methods can not be used in the current environment for network traffic anomaly detection,and it does not have the characteristics of high recognition rate and low false alarm rate.In addition,these traditional machine learning model algorithms will have some problems that are not well solved when applied to the field of network anomaly detection.First,the traditional machine learning models have poor performance,that is,the accuracy of classification is poor,especially for the multi classification problems,the effect is not very ideal.Then there is a more important point,that these machine learning methods need to do a separate step of feature extraction,which will cause inaccurate feature extraction,which leads to inaccurate classification and recognition of models.Therefore,choosing the appropriate feature combination becomes an important problem of these models.At the same time,although the recognition effect of the deep learning model has been improved,there are problems such as long training time prediction time and poor performance of the model.Moreover,the imbalance of data in open data sets will lead to the decrease of accuracy and false positivity of intrusion detection.The problem of data imbalance is also one of the difficult problems to be solved.The research in this paper is mainly aimed at the above problems.Aiming at the problems of poor classification effect of traditional machine learning and the need for artificial feature extraction,this paper proposes a network traffic anomaly detection model based on deep learning convolution neural network,which improves the recognition effect and forecasts the long training time of the deep learning model,The paper proposes a new model of traffic anomaly detection based on depth enhanced learning depth Q network,which improves the training speed and prediction speed.At the same time,aiming at the over estimation of deep Q network,the paper further proposes a traffic anomaly detection model of deep dual Q network,and also obtains better recognition effect,At the same time,the problem of data imbalance in network intrusion data set is solved,and the recognition effect is better.The model of network traffic anomaly detection proposed in this paper improves the classification effect of the model.The network traffic anomaly detection model based on deep learning is a network traffic anomaly detection model with short training time prediction time and better classification effect.At the same time,it also provides a new way of thinking for the field of network traffic anomaly detection,so it has a very high research significance.