| 摘要: | 隨著互聯網的興起，車聯網（Vehicular Ad Hoc Networks, VANET）作為智能交通系統的核心組成部分，正逐漸受到關注。利用無線通信技術和先進的車輛感知器，車聯網實現了汽車間及汽車與基礎設施間的即時數據交換，提升了交通系統的安全性和效率。然而，快速都市化導致車輛數量劇增，交通事故頻發，尤其是追撞事故。為解決此問題，本研究提出基於5G網路的車輛影像協作共享功能，通過無線通信和跳點通訊技術延伸車輛影像傳輸範圍，並使用強化學習選擇最佳傳輸對象，提高影像品質。此外，本文引入多接取邊緣計算（Multi-access Edge Computing, MEC）以應對車
載系統計算能力不足導致的延遲問題。MEC將計算資源部署在靠近用戶車輛的位置，減少數據傳輸延遲並降低網路頻寬需求。
本文通過多代理人強化學習架構，多代理人深度確定性策略梯度（Multi-Agent Deep Deterministic Policy Gradient,MADDPG），結合邊緣計算，預測最佳路徑並選擇最合適的協作車輛，實現最佳傳輸效率，提升整體系統的性能和可靠性。最後進行不同強化學習模型之間的效能比較，證明部署多代理人強化學習能使系統得到長期最大報酬。;With the rise of the IoT, Vehicular Ad Hoc Networks (VANET) have gradually gained attention as a core component of intelligent transportation systems. Utilizing wire
less communication technologies and advanced vehicle sensors, VANET enables real-time data exchange between vehicles and between vehicles and infrastructure, enhancing the safety and efficiency of transportation systems. However, rapid urbanization has led to a dramatic increase in the number of vehicles and frequent traffic accidents, particularly rear-end collisions. To address this issue, this study proposes a vehicle image cooperative sharing function based on 5G networks. By extending the range of vehicle image transmission through wireless communication and multi-hop transmission , and using reinforcement learning to select the best transmission targets, the quality of the images is improved.
In addition, this paper introduces Multi-access Edge Computing (MEC) to address the issues of insufficient computing ability and transmission delays in onboard systems.MEC deploys computing resources close to the user vehicles, reducing data transmissiondelays and less network bandwidth requirements.This paper employs a multi-agent reinforcement learning framework, Multi-Agent Deep Deterministic Policy Gradient (MADDPG), combining edge computing to predict optimal routes and select the most suitable cooperative vehicles, achieving optimal transmission efficiency and enhancing the overall system performance and reliability. Finally,
a performance comparison between different reinforcement learning models is conducted, demonstrating that deploying multi-agent reinforcement learning can achieve long-term
maximum rewards for the system. |
