Design And Implementation Of DDoS Attack Traffic Detection Algorithm Based On Deep Learning

Since the late 20 th century,the Internet has been heavily supported by technology and has grown at a rapid pace,allowing users to share files,communicate in real time and also collaborate on work in different locations by sharing computing resources.However,as the resources of the Internet have increased,this service has faced many potential attacks.These attacks can take many forms,including attacks against the physical information technology environment,exploiting vulnerabilities in applications,and using third-party applications.Distributed denial of service(DDoS)attacks,where a collection of computers is used to carry out coordinated attacks that bring down networks,are a serious problem threatening cyber security,military and medical as well as commercial aspects.Based on the theories related to DDoS attacks and deep learning,DDoS attack detection dataset(NSL-KDD,CICIDS2017)as the basis of the research,the NSL-KDD dataset is used to train the improved convolutional neural network model,which is designed to solve the problem of small sample size and few sampling features,and the model is experimentally verified to have some application value.The purpose of this paper is to solve the detection problems caused by using the traditional recurrent neural network algorithm due to the large amount of data,many sampled features and long detection time.Therefore,a sliced recurrent neural network model(SRNN)is proposed in this paper and applied to a DDoS attack detection system.The dataset used in this algorithm is CICIDS2017.The core strategy of the algorithm is to divide the DDoS dataset into several subsamples,and then a neural network model is used to process the data of each subsample in parallel.Since each level of iteration of sub-blocks in this model can be performed simultaneously,making it reliably parallel.In addition,the Adam optimization algorithm can effectively replace the traditional stochastic gradient descent algorithm,and multiple experimental results show that the training efficiency and accuracy of the SRNN model are much better than that of the RNN model.Finally,we trained neural network models by using a distributed deep learning system developed by Docker and Kubernetes,and conducted experimental comparisons.The accuracy of the SRNN(4,3)model is experimentally proven to be extremely good,and its training speed is extremely fast,which can reach 98.28% accuracy in only 422.88 seconds.