Research And Implementation Of DDoS Attack Detection Method Based On Deep Learnin

Distributed denial-of-service(DDoS)network attacks,with favorable network environments,more easily consume the resources of the attacked,and the resulting impact poses a great threat to each company or organization with network resources,especially for backbone networks.The current DDoS attack traffic and the concurrent traffic of legitimate users have converged on the characteristic information,and as the network traffic transmission speed accelerates,the existing detection methods gradually show the limitations.Therefore,under the goal of DDoS attack detection based on deep learning,the detection of DDoS attack is studied at the level of frequency domain analysis and deep learning model structure,respectively.The complete network traffic sequence can represent user behavior well,but the current DDoS attack detection methods using network traffic for feature analysis are difficult to better distinguish normal traffic from attack traffic and to ensure detection efficiency.Therefore,firstly,the network traffic is sampled regularly according to the time series and the discrete wavelet transform is applied to obtain the frequency domain information,which is combined with the statistics extracted from the original sampled signal to form a Vetter collection.Secondly,a self-encoder-based network model is proposed to enhance the hidden feature expression by improving the self-encoder network structure and adding weights to the calculation of statistics and frequency domain information in the forward propagation process;finally,an adaptive knowledge distillation model compression method is proposed to achieve effective model compression and increase the detection efficiency of the model by dynamically changing the hyperparameters through the gradient descent principle.Experiments prove that the method has good recognition effect on DDoS attacks and can obtain better detection models.The nonlinear hidden features encoded by self-encoder of a Vetter collection are used for the detection task,and the generalization ability needs to be improved.Therefore,firstly,a data reconstruction method is proposed to achieve optimal operation in terms of data by regularizing and reconstructing one-dimensional data into two-dimensional data;secondly,a fused sparse attention calculation module is proposed and the attack frequency estimates are presented and used to guide the sparse attention calculation to achieve detailed feature extraction of the data.Finally,the frequency domain information and statistics are calculated in sub-regions to reduce the interference between information.Experiments prove that the method can effectively improve the generalization ability of the model.The DDoS attack detection tool is designed and built through the research results of this thesis.The tool is built by the front-end and back-end separation technology,in which the back-end functions include listening to the network status and feeding back to the front-end display,collecting network information,discrete wavelet calculation,statistics extraction,and attack detection;the front-end functions include traffic monitoring,logging and setting interface to achieve a friendly interactive role.In the testing stage,the network topology diagram is built,and the DDoS attack and detection are simulated by two computers to test the detection efficiency and detection capability of the proposed method in this thesis.