摘要：

随着物联网和人工智能(AI)的技术发展及产品在各业务领域的推广，将边缘计算与AI模型集成融合，实现物联网智能化与计算前置化能满足更多的应用场景，但边缘代理设备通常受到硬件资源能力、性能及安全隐私等问题限制，将AI和边缘计算有效融合集成存在较大挑战。在物联网系统中，基于AI对边缘架构进行优化调整，构建具备边缘计算及AI能力的物联网新型智能框架，有效实现将边缘计算和AI集成到物联网系统中。在边端侧AI模型引导阶段，设计私有数据和公共数据的存储策略，有效提高数据安全性；在模型部署阶段，设计可配置压缩比的云端压缩、边端解压缩的部署模式，减少模型大小和传输所需数据流量，实现模型在边端侧的轻量级部署；在模型学习阶段，设计迁移学习和增量学习互补的学习方式，增加边端侧的模型训练及实用能力，提高云-边协作水平。实验结果表明，集成在边端的AI模型在资源占用率不足云模型50%情况下，准确率达到88%，同时训练时间比云模型快5倍以上。

关键词: 物联网, 边缘计算, 人工智能, 深度学习, 云边协同

Abstract:

This paper explores the integration of edge computing with Artificial Intelligence(AI) models to enhance the capabilities of the Internet of Things(IoT)and expand its application potential. However, challenges arise from limitations in edge devices, including hardware resource capacity, performance, and concerns related to security and privacy. This study addresses the effective fusion of AI and edge computing by presenting a novel IoT framework. This framework improves data security by developing storage strategies for private and public data. Additionally, it enhances model training and its practical applications at the edge through migration learning and incremental learning, thereby enhancing collaboration in the cloud environment. Concurrently, the framework achieves lightweight deployment of models on the edge by implementing cloud compression and edge decompression techniques. Despite the constrained hardware capabilities of edge devices, the proposed approach ensures high AI recognition accuracy and performance. Experimental results demonstrate that the resource usage of edge-integrated AI model is less than 50% compared to cloud-based models, with edge-based AI achieving an accuracy rate of up to 88%. Moreover, its training time is more than five times faster than that of cloud-based models.

Key words: Internet of Things(IoT), edge calculation, Artificial Intelligence(AI), deep learning, cloud side collaboration