作为大数据时代的一个极其重要源头，来自5G移动互联网与智能物联网的移动大数据，将在智慧城市、健康计算、环境监测等众多领域得到广泛应用。新兴移动大数据应用具有计算密集、动态性强、要求实时响应等鲜明特点，对传统云计算提出了巨大挑战。为此，本项目研究端边云多层次融合高效边缘计算机制，以解决信息获取、多元协同、高效调度关键问题为抓手，有机融合智能终端、边缘节点以及云端平台的差异化优势，来高效支撑移动大数据应用的服务需求。首先，项目提出了能效优先的资源信息感知机制和数据驱动的任务建模方法，为用户边缘计算提供高质量决策信息。其次，项目设计了传算协同的终端任务卸载优化算法，高效聚合多层次通信与计算异构资源协同的优势。项目进一步提出了基于分布式深度强化学习的边云协作智能调度机制，采取人工智能赋能策略驱动众多边云节点高效协作联合调度。最后，项目通过设计实现多层次边缘计算原型系统来检验上述算法和机制的性能。

As a highly critical data source in the era of big data, mobile big data which is generated from 5G Mobile Internet and Intelligent Internet of Things will have wide applications in many fields, including smart city, health computing and environmental monitoring. Mobile big data applications possess many distinct characteristics such as being computationally intensive, highly dynamic and demanding real-time response, which raise significant challenges for the traditional cloud computing architecture. For well supporting the service demands of mobile big data applications, in this proposal we study the efficient edge computing mechanisms with multi-layer synergy of device-edge-cloud, and address the key issues of system information gathering, heterogeneous resource cooperation and efficient task scheduling by amalgamating the differential advantages of end devices, edge nodes and cloud platforms. First, we propose an energy-efficient resource information sensing mechanism and a data-driven task modeling method, to provide high-quality decision information for optimizing user’s edge computing. Second, by fully leveraging the advantage of multi-layer heterogeneous resource cooperation, we design a device task offloading optimization algorithm based on the synergy of communication and computation. We further propose a distributed deep reinforcement learning based intelligent scheduling mechanism, aiming at utilizing powerful AI techniques for achieving efficient collaborative scheduling among the multitudinous edge and cloud nodes. Finally, we evaluate the performance of the proposed algorithms and mechanisms above through designing and implementing a multi-layer edge computing experimental system prototype.