多云架构下的人工智能开发平台设计  

Design of Artificial Intelligence Development Platform Under Multi-cloud Architecture

作　　者：雷鸣 姜罕盛 勾志竟 郭阳 Lei Ming;Jiang Hansheng;Gou Zhijing;Guo Yang(Tianjin Meteorological Information Center,Tianjin 300074,China)

机构地区：[1]天津市气象信息中心,天津300074

出　　处：《单片机与嵌入式系统应用》2022年第7期14-16,29,共4页Microcontrollers & Embedded Systems

基　　金：国家自然科学基金(41575156);天津市科技局科技重大专项与工程(18ZXAQSF00130);天津市气象局科研项目(201914ybxm12)。

摘　　要：气象行业数据具有一定程度的保密性、复杂性和安全性,在开放性云资源的使用上面临风险,且单一云架构无法有效满足气象行业的数据需求。针对这一问题,本文提出了一种多云架构下的人工智能开发平台设计方案。实验结果表明,该方案可以形成优势互补,满足不同情况的用户需求,在天河云资源上构建的系统相比在CPU计算资源上的计算速度提升了16.77倍,加速效果明显,能够有效满足气象行业对AI算力资源的需求。Meteorological industry data has a certain degree of confidentiality,complexity and security.It will face risks in the use of open cloud resources,and a single cloud architecture can not effectively meet the data needs of the meteorological industry.To solve this problem,a design scheme of artificial intelligence development platform under multi-cloud architecture is proposed.The experiment results show that this scheme can form complementary advantages and meet the needs of users in different situations.The computing speed of the system built on Tianhe cloud resources is 16.77 times higher than that on CPU computing resources.The acceleration effect is obvious,and can effectively meet the needs of meteorological industry for AI computing resources.

关 键 词：深度学习 AI平台 多云架构 ModelArts 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]