Optimization of Road Condition Decision at Intersection of V2X Control System based on Federated Learning and Edge Computing

DOI: https://doi.org/10.62517/jike.202604222

Author(s)

Jiawei Tian

Affiliation(s)

School of Automotive Engineering, Harbin Institute of Technology, Vehicle Engineering, Weihai, China

Abstract

As a kind of vehicular wireless communication technology, the Internet of vehicles (V2X) is a new generation of information and communication technology that connects vehicles with all objects on the road, where V stands for vehicles and X stands for objects that interact with the vehicle. At present, X mainly includes vehicles, people, traffic road infrastructure and network. With the help of modern wireless communication technology, the Internet of vehicles technology can realize the interaction between vehicles and between vehicles and the outside world, and can fully automatically remind drivers of possible safety risks, so as to effectively avoid traffic accidents. [1] With the rapid development of Internet of vehicles (IoV) technology, autonomous driving is evolving from single-vehicle intelligence to multi-agent vehicle-road collaboration mode, and its potential in improving road safety and traffic efficiency has attracted much attention. Current V2X applications, however, still faces significant challenges: network fluctuation result in higher data transmission delay, the cloud centralized computing model of great communication and calculate power burden; The traditional path planning model based on distributed learning is prone to make redundant or delayed decisions in dynamic scenarios such as complex intersections due to the large number of parameters and poor scene adaptability, and it is difficult to make the optimal path planning. Aiming at the problem of the traditional distributed learning, this paper puts forward a kind of edge fusion calculation and federal study of the collaborative decision-making optimization framework. Edge computing reduces the end-to-end transmission delay by offloading data processing to road side units or vehicle terminals. Federated learning supports local training of multi-vehicle data and global model aggregation, which reduces redundant gradient calculation while protecting privacy and enables the control system to choose the optimal route. Based on this framework, this paper adopts CarSim associated with Simulink simulation platform, to build a typical urban intersection set, in view of the automatic vehicle decision-making process is optimized. In the experiment, the key indicators of the optimized vehicle such as driving curve, speed change rate, acceleration change rate (ride comfort) are analyzed, and the advantages of the proposed method are verified. Results show that the research for the lightweight, highly efficient of the autopilot system, privacy decision-making provides a new train of thought, has the important value of engineering application.

Keywords

Automatic Driving; Co-Simulation; Federated Learning; Experimental Analysis

References

[1] Yang Yitong, Wu Jingjing, Zhang Lili. Design and implementation of V2X based Internet of Vehicles experimental system [J]. Modern Electronics Technology,2024,47(24):33-37. [2] WANG Jianzong, KONG Lingwei, HUANG Zhangcheng, et al. A Survey of Federated Learning Algorithms [J]. Big Data,2020,6(06):64-82. [3] HU Bingxu. Research on Intelligent Vehicle Obstacle Avoidance Algorithm Based on Improved Federated Learning [D]. University of Electronic Science and Technology of China,2023.[C] QIU Fan. Research on resource allocation of C-V2X Internet of Vehicles based on multi-agent federated learning [D]. South Central University for Nationalities,2023. [4] SUN Lin. Research on Efficient Federated Learning Algorithm for Autonomous Driving [D]. Xidian University,2024. [5] QIU Fan. Research on Resource allocation of C-V2X Internet of Vehicles Based on Multi-agent Federated Learning [D]. South-central University for Nationalities,2023. [6] SHI Weisong, Sun Hui, Cao Jie, et al. Edge computing: a new computing model in the era of Internet of Everything [J]. Journal of Computer Research and Development,2017,54(05):907-924. [7] Wang Y, Wang Y, et al. A novel approach to multi-agent cooperative decision making based on Federated Reinforcement Learning [J]. Beijing University of Posts and Telecommunications,2024. (in Chinese) [8] WU Wentao. Research on Autonomous Driving Strategy of Internet of Vehicles Based on Federated Learning [D]. Shanxi University,2023.