随着5G技术的逐步推广，越来越多的智能设备将融入到城市物联网中，为智能交通服务提供更多样的感知能力与更充沛的分布式算力，那么如何有效利用这些新能力，在保证数据隐私与安全的前提下，训练出更准确的机器学习模型，为服务提供智能化核心，成为重要的研究方向。又因集中式机器学习无法有效解决由隐私保护而产生的数据孤岛问题，所以去中心化的联邦学习模型逐渐成为研究的重点。总观当前对联邦学习模型的研究，多面向同步且同质的分布式环境，尚不能有效适配因5G技术的应用所引入的异步且异构的运行环境，所以需要针对该设定对联邦学习模型做更深层次的探索。为此，本项目将研究面向5G多样异构智能端协同的异步联邦学习模型，通过层次化融合边缘端、雾端与云端资源、阶段式训练联邦模型，实现资源的动态结构化管理与联邦模型的快速准确训练；同时，基于所研究的模型，将对个性化出行服务进行原型的设计与实现，为模型的实际应用提供范例。

With the gradual utilization of 5G technology, more and more smart devices will be integrated into the Internet of Things (IoT) of the city to provide more diverse sensing capabilities and more abundant distributed computing power for intelligent transportation services. Therefore, how to effectively use these new capabilities and, in the meanwhile, ensure data privacy and security to train more accurate machine learning models as the intelligent cores of diverse services has become an important research direction. However, as the centralized machine learning approach cannot address the data isolation issue caused by the enforcement of data privacy protection regulations, the decentralized federal learning is introduced and drawing more attention from the research community as the focus of future research. Moreover, since the existing research on federated learning primarily target synchronous and homogeneous distributed environments, which cannot effectively overcome the challenges introduced by the application of 5G technology, the research on federated learning model under the constraint of heterogeneity and asynchrony is needed as a new topic. To address this emerging requirement, this project is studying an asynchronous federation learning model in collaborating diverse and heterogeneous intelligent ends towards 5G with capabilities of hierarchically orchestrating edge, fog and cloud resources, and rapidly learning federated model in multiple phases to achieve optimal management of dynamically structured resources and fast training of more accurate federated model; at the same time, based on the research model, a personalized travel assistant will be prototyped to provide a guideline of using the model in supporting the evolution of smart mobility services.