Cloud-edge-device Integration Based On Heterogeneous Computing AIoT Architecture Design And Application

The digital economy has become the main engine driving China’s economic growth,and AIo T(Artificial Intelligence & Internet of Things),as an emerging technology in the wave of digital economy development,is playing a huge role.AIo T is a form of Io T application and research in which artificial intelligence and the Internet of Things are deeply integrated.Io T can be regarded as the infrastructure layer of AIo T,which collects and provides a large amount of sensor data,while AI helps Io T realize intelligence connected tools.As the concept of AIo T is put forward,traditional Io T is rapidly transforming into AIo T,and this change also brings unprecedented difficulties and challenges.Due to the explosive growth of edge terminal devices,massive amounts of sensor data are generated.We need to intelligently identify and mine the sensor data,and store massive amounts of AI training data and data after intelligent identification and analysis.This poses huge challenges to network bandwidth,computing power and storage,and the traditional Io T architecture is difficult to meet the current needs.This paper integrates technologies and theories such as heterogeneous computing,distributed technology,and cloud-edge collaboration,and designs and implements a cloud-edge-device fusion AIo T architecture based on heterogeneous computing.Due to the diversification of AIo T scenarios,a single computing architecture cannot solve the bottleneck of computing power.This paper designs and implements an algorithm engine that supports heterogeneous computing,which can use GPU multi-core parallel computing to accelerate the algorithm.Some KNN algorithms implement parallelized reconstruction to speed up computation and support dynamic switching of algorithm libraries.The architecture of this paper is layered into the AIo T edge layer and the AIo T cloud core layer.At the edge layer,a high-concurrency and high-availability intelligent Io T gateway software is designed for the problems of insufficient bandwidth and low processing efficiency caused by massive device access.The edge-layer gateway implements heterogeneous data access by encapsulating multiple communication buses,Define parsing rule templates to realize data processing and data storage through database technology.In addition,a low-latency edge-cloud collaborative AIo T solution is proposed for the delay-sensitive problem and insufficient computing power of AI processing.The architecture integrates edge nodes and cloud core nodes through the registration center,designs a dynamic scheduling scheme according to the transmission delay between nodes and node load,etc.,and caches it to edge nodes to assist in selecting the optimal node to realize computing offloading tasks.The cloud core layer also provides services such as data analysis and data persistence.In addition,it provides API capabilities to the outside world,and realizes the ability to issue commands through a common interface.It is a complete solution for AIo T applications.The system designed in this paper also proves its feasibility and reliability through landing applications.It has been installed and operated stably in school dormitories,rental houses,charging piles and other scenarios.