Design Of A Deep Learning Development Platform For Meteorological Researc

With the rapid development of the branch of deep learning in the field of artificial intelligence,the deep learning model for spatiotemporal series prediction for the meteorological industry is under hot research and application.In recent years,major companies at home and abroad,as well as universities,are integrating the underlying computing resources,focusing on increasing investment in deep learning.In order to simplify the calculation cycle of deep learning and reduce the entry threshold of users,this paper aims at the full cycle of deep learning in meteorology and the high availability of the platform,and builds a deep learning platform for meteorological research,aiming at improving the efficiency of users’ deep learning.The main research work is as follows:(1)Using Docker containerization technology to integrate meteorological deep learning and use a wide range of deep learning framework,provide effective solutions for users in terms of convenience of environment configuration and security of environment isolation,and quickly deploy some services required by the platform through Docker.(2)Build a full-cycle in-depth learning platform for meteorological in-depth learning,involving a multi-functional platform for meteorological in-depth learning dataset management,model training,and model deployment,and provide users with a concise and clear visual interface,and treat users with a low degree of professionalism.(3)Build a relatively comprehensive platform resource management system based on Kubernetes container orchestration technology.Users can run customized in-depth learning tasks in the cluster without caring about the underlying resources.By integrating multiple cloud native technology solutions,create a reasonable and complete technology stack to achieve the integrated allocation of computing resources and integrated management of platform resources.In addition,in order to verify the functional and non-functional implementation of the platform,this paper analyzes the deployment test of the platform.The design and application of this platform can meet the basic requirements of users for in-depth meteorological learning,and reduce the training deployment cost and entry threshold of users.It has contributed its own strength to the development of meteorological information industry through the integration and configuration of underlying resources.