智能网联汽车已成为汽车产业发展方向，但目前在大规模应用上还面临着严峻的问题。单车感知、计算能力的局限性以及多车之间行为存在难以协调的潜在冲突和博弈是两大关键问题，智能网联汽车云控系统是突破这些局限的有力方法。本文讨论了在云控系统架构下计算任务的工作逻辑，总结了云控系统关键需求以及计算任务的数学建模方式，并建立了云控系统计算调度优化方法，同时开展了云控系统原型的开发和应用。 具体地，对于计算调度优化问题，考虑车辆移动性以及智能网联汽车云控系统动态特性带来的服务迁移的因素，以动态过程中的系统成本与性能为目标构建了长时域优化问题。利用李雅普诺夫优化方法，将长时域优化问题在时间域内进行解耦，整形为实时优化问题。针对优化决策中多车之间争用云端资源的问题，设计了计算资源租赁的机制，构建了多车之间的动态古诺博弈关系。基于下降法设计了分布式求解算法，通过迭代，使得系统达到纳什均衡状态，实现多车之间资源的优化分配。从而提出了适合智能网联汽车云控系统环境的分布式实时计算调度优化算法，实现了长时域以及全局协同的优化。 基于SUMO和OMNet++构建了智能网联汽车云控系统仿真平台，在SUMO中构建了道路交通模型，在OMNet++中开发了相关的网络协议与云控组件模型，建立了仿真测试场景。使用Veins框架，实现了SUMO和OMNet++的联合仿真，开发了自动化测试工具，进行大规模仿真的批处理。仿真测试结果验证了优化算法对于云控系统性能的优化效果，并以自适应巡航、车辆队列控制场景为例，验证了云控平台对于车辆应用的支持作用。 为了更深入地测试和验证云控系统的性能，本文搭建了智能网联汽车测试场与云控系统原型，部署了基于LTE-V协议的车载与路侧单元、路侧传感器、云控平台的硬件设施。在软件层面利用容器技术Docker与集群管理技术Kubernetes，设计与实现了云控系统的软件技术架构。以车端与路侧融合感知应用为例进行了容器环境下的应用开发与测试验证，证明了基于容器技术的云控平台能够满足该应用的功能与性能的需求。
The development of Intelligent Connected Vehicles has become the strategic direction of the global automotive industry, but there are still severe problems in its applications. The limitations of computing and perception capabilities, as well as conflicts and competitions among vehicles, are two key issues. Intelligent Connected Vehicles’ Cloud Control System is a promising way to break through these limitations. This paper discusses the working logic, system requirements, and mathematical modeling methods of Cloud Control System, and then proposes optimization strategies for Cloud Control System in terms of computation task scheduling. Specifically, for the factors of service migration caused by vehicle mobility and the dynamic characteristics of Cloud Control System, a long-term optimization problem is constructed, and the cost and performance in the dynamic process are the optimization goals. Using Lyapunov Optimization, the long-time domain optimization problem is decoupled in the time domain and becomes a real-time optimization problem. Considering the problem of multi-vehicle contention for cloud resources in optimization decision-making, the mechanism design of computing resource leasing is adopted to construct a dynamic Cournot game relationship among multiple vehicles. Based on the descent method, a distributed iteration algorithm is designed. Through iteration, the system achieves the Nash equilibrium state and achieves the approximate optimal resource allocation among multiple vehicles. In this way, the long-term domain and global collaborative optimization of computing task scheduling are realized, and the strategy of computation scheduling in the environment of Intelligent Connected vehicles’ Cloud Control System is proposed. Based on SUMO and OMNet++, a simulation platform for Intelligent Connected Vehicles’ Cloud Control System is built. Road and traffic models are constructed in SUMO, and network protocols and cloud control component models are developed in OMNet++, and then test scenarios are established. Veins framework is used to realize the joint simulation of SUMO and OMNet++, and then an automated test tool for batch processing of large-scale simulations is developed. The simulation results verify the optimization effect of the optimization algorithm on the performance of the Cloud Control System. Taking Adaptive Cruise Control, vehicle queue control as examples. This paper verifies the support effect of the Cloud Control System on vehicles' applications. To test and verify the feasibility and performance of the Cloud Control System in more depth, this paper builds an Intelligent Connected Vehicle test field and a prototype Cloud Control System. The test filed contains RSU and OBU based on the LTE-V protocol, several roadside sensors, and infrastructure of Cloud Control System. Using container technology Docker and cluster management technology Kubernetes, This paper designs and implement the software architecture of Cloud Control System. Taking the car-side and roadside fusion perception application as an example, the application development and verification test in the container environment have been carried out, which proves that Cloud Control System based on container technology can meet the requirements of function and performance.
