Research And Implementation Of Intelligent Autonomous PHM System For Cloud Platform

Cloud platform is an essential new infrastructure in the digital industry.As various fields are now competing to embrace cloud computing technology,cloud platforms have a wider application space.However,the complexity and integration of cloud platforms' functions are increasing in order to better serve the increasingly demanding and changing application scenarios.This has led to an increasing probability of server failures under the cloud platform,and further led to an increase in cloud platform maintenance costs.Traditional cloud platform operations and maintenance work requires artificial analysis of a large number of indicators to determine the health of the server,but this approach is very inefficient and extremely dependent on the experience of relevant experts.Moreover,when a server fails,the response to the failure can only be passive.To address these issues,this paper applies prognostics and health management(PHM)technologies to cloud platforms,thus improving the status quo of lagging server operation and maintenance management on cloud platforms.The contributions of this paper are as follows.(1)The related theories and methods of PHM are studied.For the differences between the cloud platform scenario and the traditional application scenario of PHM,this paper selects a suitable methodological theory and implementation route for the implementation of the target system.(2)The principle and application of variational autoencoder are studied.To address the challenge of health index construction required by the data-driven PHM approach,this paper investigates the learning ability of the variational autoencoder for normal patterns in time-series data.Then the ability of the model to detect abnormal patterns of indicators is verified on collected data.Finally,a index indicating the health of the cloud platform server is obtained indirectly using the above capability.(3)The overall architecture of the PHM system is designed.This paper analyzes the current pain points of cloud platform server operation and maintenance management,and forms specific requirements for data collection,real-time status monitoring,data visualiza-tion and health management.Then the overall design of the target system's hierarchical,functional and logical architectures is carried out according to the requirements.(4)The PHM system was implemented and tested.In this paper,the mainstream technology framework is chosen to design and implement the target system,and the design logic and implementation steps are given.Finally,the test shows that the target system of this paper can work properly.The proposed PHM system in this paper realizes real-time anomaly detection and overall failure prediction of the server,which effectively improves the cloud platform's ability to respond to failure and effectively guarantees the long-term availability of the server.At the same time,the reliability of the cloud platform and the automation and intelligence level of platform management are improved.