| In recent years,the contradiction between the increasing amount of data and the limited computing power of individual computers has given rise to the concept of cloud computing.Relying on the powerful computing power and convenient distributed service capability of the cloud environment,cloud computing has been widely used in various key areas of society,such as transportation,finance,and healthcare.However,the rich computing and storage resources behind cloud computing are easy targets for hackers,and how to secure the cloud environment has been a key research problem in academia and industry.Intrusion detection is an important technology that can effectively detect and secure the network.Considering that the cloud environment has a larger amount of data than the traditional network environment,and requires higher timeliness,efficiency and accuracy of detection,how to combine intrusion detection with the cloud computing environment organically to ensure the network security of the cloud environment is an urgent problem nowadays.Considering that artificial intelligence technology can effectively deal with massive data,many contemporary research works have applied artificial intelligence technology to intrusion detection in cloud platforms and achieved certain results.Inspired by these works,this paper also combines artificial intelligence technology with cloud platform intrusion detection accordingly.Through the organic combination of these two technologies,this paper makes corresponding improvements on the traditional intrusion detection mode and designs an intelligent intrusion detection method with high detection efficiency,high accuracy,robustness and security that is more applicable to the cloud environment,which effectively improves the security protection capability of the cloud platform.The work in this paper focuses on the following three points respectively.First,this paper deeply analyzes the current security requirements of the cloud platform,studies the current artificial intelligence-based intrusion detection technology,and presents the defects and problems of the current technology.Second,to address the problems of feature redundancy and excessive computational overhead brought by high-dimensional data faced by machine learning-based intrusion detection technology,this paper designs an improved Genetic Algorithm(GA)to perform feature selection,and then realize data dimensionality reduction,which effectively improves the efficiency and accuracy of machine learning-based intrusion detection technology.Specifically,this paper adds normalization and nonlinear variation operations to the Fitness function of the GA algorithm to improve the robustness of the Fitnees function,and designs the individual variation probability and variation direction based on the correlation of the data features in the variation operation of the GA algorithm to improve the effect of the variation operation.Experiments show that the data after dimensionality reduction by the improved GA algorithm in this paper can be adapted to the intrusion detection algorithm and its usability is high.Finally,to address the problem that the intrusion detection algorithm based on deep learning is easily disturbed by small perturbations and vulnerable to security attacks,this paper investigates the optimal ε value and the optimal proportion of adversarial samples to improve the reliability and security of the intrusion detection algorithm based on deep learning,starting from the adversarial attacks in the adversarial defense.Meanwhile,for the Free robust training method,this paper explores the impact of different adversarial sample attacks on the robustness of the final model on multiple intrusion detection datasets,and accordingly proposes the MI-Free robust training method that can greatly improve the robustness of the model.The experimental results show that the MI-FGSMbased Free robustness training method MI-Free can effectively improve the robustness of the intrusion detection model with high usability. |
PDF Full Text Request