Artificial Intelligence-based IaaS Cloud Adaptive Anomaly Detection Technology

With the advent of the information age,Cloud Computing has been widely used in various fields.However,while the Cloud Computing system becomes more and more large and complex,the frequency of various anomalies faced by the Cloud Computing system is also increasing.Cloud service providers need to constantly improve the security and reliability requirements of the whole system.IaaS is one of the most popular services provided by cloud service providers today.It provides virtual machines of different specifications for users to rent.Cloud service providers can also find and solve anomalies in time through real-time monitoring and detection of virtual machines,so as to improve the security of cloud computing system.However,the performance indicators of virtual machines in the cloud are highly complex and large in number.The complexity of the operation habits of different virtual machine users in the cloud makes the corresponding monitoring indicators greatly different.The traditional manual analysis method is more and more difficult to adapt to the current large-scale,complex and heterogeneous systems,and it's easy to ignore some problems,such as memory anomalies.Based on CNN model and Reinforcement Learning theory,the thesis proposes an Artificial intelligence-based adaptive anomaly detection technology to deal with memory anomalies in IaaS cloud virtual machine.(1)An Artificial intelligence-based adaptive anomaly detection framework is proposed to deal with memory anomalies.The overall framework is composed of three levels of modules: anomaly detection module,transfer training module and central scheduling module.Taking the four key performance indexes of CPU,memory,hard disk and network as the Input data source,the adaptive detection of memory anomalies in different virtual machines in IaaS cloud is realized.(2)Zoom-CNN anomaly detection model based on TCN and CNN model is proposed for anomaly determination in the target virtual machine,which effectively reduces the monitoring requirements of the system for the target virtual machine,improves the monitoring and detection quality,and reduces the operation and maintenance resource consumption of the system.(3)A transfer training module based on the principle of Transfer Learning is proposed to reset or update the parameters of the anomaly detection model,so that the model can better match the target virtual machine,and improve the anomaly detection accuracy of the anomaly detection model in the face of strange virtual machine.(4)The central scheduling module is constructed based on DQN algorithm.The Reward function is constructed by the virtual machine state,anomaly detection result and loss value feed back by the transfer training module.Through the interaction and reward with the transfer training module,the automatic control of the transfer training module is realized,so that the anomaly detection model can achieve a better matching with the target virtual machine,and the overall framework can be adjusted adaptive.