Jisuanji kexue (Sep 2022)

Functional Architecture to Intelligent Computing Power Network

  • HU Yu-jiao, JIA Qing-min, SUN Qing-shuang, XIE Ren-chao, HUANG Tao

DOI
https://doi.org/10.11896/jsjkx.220500222
Journal volume & issue
Vol. 49, no. 9
pp. 249 – 259

Abstract

Read online

The computing power network as a new research fieldis in urgent need to improve intelligence and provide on-demand services.To solve the problems,an intelligent computing power network that combines cloud-edge-terminal computing resource,communication resource,and AI approaches together is proposed.At the same time,the system-level intelligence would be built from two aspects of endogenous intelligence and application intelligence.The endogenous intelligence refers to the abilities of self-perception,self-adaptation,self-decision and self-learning to make the computing power network ensure the accurate operation of the system.The application intelligence refers to the abilities in resource deployment,business arrangement and cognition,so that the computing power network could enhance the adaptability of the businesses.Further,a functional architecture is constructed,which would gradually build the endogenous intelligence and application intelligence in the layers of basic resource,resource mana-gement,business choreography,operation service and optimization.Finally,three important scenarios are selected from two domains,i.e.workshop logistics and quality control based on machine vision in the smart manufacturing,road and community monitoring in the intelligent security.Three groups of simulations are designed based on the scenarios,respectively.Experimental results show that while adopting the intelligent computing power network with endogenous intelligent and application intelligence,in the scenario of workshop logistics,the performance improvement is related with the scale of the scenarios,the planning time could be improved by approximately 2～50 times and the planning result could be improved by about 2～5 times.In the scenario of quality control based on machine vision,the deploymentcost of computing equipment would reduce to 1/5 of the original and the detection accuracy could improve by about 4.5%.In the scenario of road and community monitoring,the deployment cost of computing equipment could reduceto 1/10 of the original.

Keywords