车联网环境下数据驱动的云服务访问建模及任务优化策略研究

Vehicular Cloud Computing (VCC) is a frontier technology which combines Vehicular Ad hoc Networks (VANETs) and cloud computing. In VCC environment, users can overcome many disadvantages of vehicular terminals, which includes limited computing power and small storage capacity, by offloading the computing and storage task to the cloud service center. .With the increasingly enrichment of traffic application, vehicular users’ requirements for the cloud service is growing greatly. However, vehicular cloud service access is restricted by multiple network environment factors due to VANETs complex dynamic properties. Therefore, the deployment of cloud computing in VANETs faces huge challenges. Existing service optimization strategies usually design optimization algorithm by establish priori model of specific network parameters. This modeling procedure can not reflect the impact of complexity properties on cloud service access. And it will leads that service response lags behind the variance of service request. Therefore, the key to solve the problem is to establish a dynamic model of service access which can reflect the complex characteristics of VANETs. This project will utilize transportation data set and introduce data driven modeling ideology to modeling procedure of VCC service. Based on it, we will stick to cloud service access procedure to study the discrete-continue modeling method and its optimization scheme based on group method of data handling neural network. And we will apply the modeling method to service access, scheduling and delivery process, establish the dynamic continuity model of related network parameters, and finally, construct the corresponding dynamic optimization strategy. In order to verify the feasibility of proposed models and methods, we will establish physical testing platform and transportation-network coupling simulation platform by utilizing transportation history data set. These testing systems will be used to overall evaluating the above proposed models and methods from different granularity. The research works of the project will provide theories and solutions for vehicular cloud service accessing model and optimization.

车联网云计算是实现车载应用服务的重要技术方向，而车联网复杂动态特性导致云服务访问效能受多种网络环境因素制约，给访问优化带来巨大挑战。现有的服务访问优化策略通常针对特定网络参数建立先验模型设计优化方案，无法体现车联网动态特性对访问优化过程的影响，导致服务响应滞后于服务请求的变化。因此，建立能够综合反映车联网复杂特性的服务访问动态模型，是解决问题的关键。本项目将根据交通离散数据，将数据驱动的连续性建模思想引入到车联网云服务访问建模过程，拟研究针对云服务访问过程的，基于数据处理分组神经网络方法的离散-连续建模策略及其优化方案，并将其应用于服务访问的接入、调度和交付环节，建立相关网络参数的动态连续性模型，进而构建相应的动态优化策略。为评估模型方案的可行性，项目将通过物理验证平台和基于交通数据集的网络耦合仿真平台，从不同层面对提出的模型方案进行综合评估，为车联网云服务建模和优化提供理论依据和方案。

在车联网应用日趋多样的情况下，在车载环境下部署云计算服务是实现智能交通系统的重要发展趋势。在本项目的研究中，通过对车联网信道状态、服务需求与通信资源的分析，深入研究了车联网数据传输调度算法、控制信道协议构建及网络安全问题。项目主要内容包括：（1）基于多信道的车联网数据传输调度，提出基于调度冲突图和冲突矩阵的全新多信道车联网协作传输调度方案来解决任务调度和多信道分配问题。（2）针对网络控制信道协议构建问题，通过分析网络链路对控制层创建过程的影响，从控制拓扑的冗余度出发，提出一种具有较好的时间开销和扩展性的分布式控制信道构建协议。（3）针对通信频谱资源稀缺问题，引入认知无线电技术提高车载网络的频谱利用率，通过对频谱状态、车辆缓存和移动性、各类传输模式以及服务质量的综合分析，提出基于深度Q学习的数据传输调度方案。（4）针对车联网中的安全和隐私问题，在物理层上，引入人工噪声技术分析了基于协作通信的非正交多址技术的通信安全性能并提出改善安全性能的终极选择方案，在网络层上提出了基于信任推导模型的多播路由协议来提高网络抗攻击性能。项目的研究成果有助于完善车联网云服务访问控制的理论体系，对提高车联网服务质量、推动智能交通系统的发展有重要意义。在本项目的支持下，共发表期刊论文19篇，会议论文1篇，联合培养毕业1名博士研究生，培养毕业4名研究生。