車載網路下基於 Stackelberg 賽局和多代理人強化學習之中繼傳輸群組建立及即時影像分享;Using Stackelberg Game and Multi-Agent Reinforcement Learning to Self-Organize Relaying Groups for Real-Time Video Sharing in Vehicular Networks

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/92340

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/92340

Title:	車載網路下基於 Stackelberg 賽局和多代理人強化學習之中繼傳輸群組建立及即時影像分享;Using Stackelberg Game and Multi-Agent Reinforcement Learning to Self-Organize Relaying Groups for Real-Time Video Sharing in Vehicular Networks
Authors:	朱育成;Chu, Yu-Cheng
Contributors:	通訊工程學系
Keywords:	仿真環境模擬;邊緣計算;車聯網;賽局理論;多代理人強化學習;車載自組織網路;Simulation environment modeling;Edge computing;Vehicular networks;Game theory;Multi-agent reinforcement learning;Vehicular self-organizing network
Date:	2023-08-14
Issue Date:	2023-10-04 15:27:11 (UTC+8)
Publisher:	國立中央大學
Abstract:	由於快速的城市化導致交通路況變得更不穩定，若是後方車輛無法得知前方的路況為何，當前方有事故發生或是異常狀況，將會導致反應不及發生追撞事件，造成嚴重的交通事故與安全問題。因此車輛間畫面的協作共享將會成為一項重要的議題，隨著 5G 和人工智慧的蓬勃發展，不但能利用無線通訊讓車載裝置之間進行快速的溝通，也能針對所收集到的數據進行分析部屬。有鑑於此，本研究首先使用車載仿真模擬器進行真實環境的建模，接著利用賽局理論對車載環境進行詳細的描述與定義，然後將其集成至多代理人強化學習模型，並採用 MADDPG 模型解決此問題，以挑選擁有最低延遲、最高數據傳輸率的最佳傳輸路徑，最終將車輛組成自組織網路以實現畫面傳輸共享。在分析方面，本研究針對不同的車載訊息傳輸方式、車載間跳點裝置的最大數，皆有進行評估比較，並比較了多代理人與單代理人強化學習之間的評估，實驗結果表明，部屬多代理人強化學習能使車載傳遞訊息時的效果更好，有較高的效能。最終本研究將針對傳輸延遲、數據傳輸率、功耗等三項指標進行不同模型之間的評估分析。;Due to rapid urbanization, traffic conditions have become increasingly unpredictable. In the scenarios of neighbor vehicles crowds, vehicles in the rear are unaware of the current road conditions ahead. Accidents or abnormal situations occur in the front can lead to delayed reactions and rear-end collisions, this phenomenon which results in severe traffic accidents and safety concerns. Collaborative sharing of visual information among vehicles becomes an important issue. With the rapid development of 5G and artificial intelligence, not only can wireless communications be utilized for fast data transmissions between in vehicle devices, but the data collected can also be analyzed and deployed. Hence, the study in this thesis first utilizes a vehicular simulation emulator to model real-world environments. Subsequently, the game theory is employed to provide a detailed description and definition of the vehicular environment. Both of the above two efforts are then integrated into a multi-agent reinforcement learning model, using the Multi-Agent Deep Deterministic Policy Gradient（MADDPG）approach. The objective is to select the optimal transmission path with the lowest latency and highest data transmission rate, thereby enabling vehicles to form a self-organizing network for video transmission and sharing. This study evaluates and compares different vehicular information transmission methods and the maximum number of hop devices between vehicles. In addition, this study compares the evaluations between multi-agent and single-agent reinforcement learning approaches. Experimental results demonstrate that deploying multi-agent reinforcement learning yields better performance and higher efficiency in vehicular message transmission. Finally, this study conducts evaluation and analysis among different models based on three metrics: transmission latency, data transmission rate, and power consumption.
Appears in Collections:	[Graduate Institute of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	64	View/Open

社群 sharing

Loading...