Research On The Anomaly Detection Technologies For Cloud Computing Environment

With the the rapid development of information technology and the booming of new ideas and methods such as distributed technology,big data,virtualization and so on,the cloud computing technology has come into being.Over the years,cloud computing has been applied to more fields,and its development is more and more rapid and powerful.The operating mode in the cloud computing has been greatly improved than before,makes the cloud computing environment relatively high utilization and availability.However,due to the massive scale and the complexity,the cloud computing is vulnerable to the anomaly situation or the malicious attack,which poses a serious threat to the security of cloud computing environment.Therefore,anomaly detection technology in cloud computing has become a research hotspot in this field.In this paper,our research mainly focus on the network anomaly which is faced on the security of the cloud computing environment network.The key point is to improve the availability of cloud computing environment.The main work of this paper is as follows:Firstly,for massive,complex,and unbalanced data in cloud environments,a method based on impact compensation factors is proposed.The differential equalization method introduces the impact compensation factors for small sample sizes and increases their impact on the entire model so that their characteristics are not submerged by large categories.Secondly,simulated annealing algorithm is introduced according to the BP neural network learning easy to fall into local minimum problem.The problem of slow convergence and easy oscillation at low temperature in simulated annealing is analyzed.A modified simulated annealing algorithm for two-stage optimization is proposed.In the two-stage,different control parameters are set respectively,which can guarantee the characteristics of global optimization and reduce the occurrence of oscillation.Thirdly,this paper analyzes the characteristics of BP neural network gradient descent method.Moreover,we propose an improved BP neural network algorithm based on cross-grouping,which aims at the characteristics of the complex training process and the correlation between samples.In addition,the dynamic adaptive learning rate is set up for the neural network to solve the problem of slow convergence speed according to the change of objective function.The learning rate is adjusted dynamically to adapt to the optimization process according to the change of objective function,which can improve the convergence speed.Our work can effectively improve the performance of the algorithm by combining the above algorithms.Fourthly,this paper design and implement a novel system model based on the three-layer architecture,which is "data processing-anomaly detection-exception response".We describe the characteristics of each functional module in detail.Based on this model,we can detect anormaly behavior accurately,and improve the safety of cloud computing environment.Finally,we run the experiment and evaluate the performance of the system.The experiment results show that our algorithms not only improve the detection rate,reduce the false detection rate,accelerate the learning speed,but also play a significant compensation role for small types.It can effectively solve the problems existing in security of cloud computing environment.