DC 欄位 |
值 |
語言 |
DC.contributor | 通訊工程學系 | zh_TW |
DC.creator | 徐偉銘 | zh_TW |
DC.creator | Wei-Ming XU | en_US |
dc.date.accessioned | 2023-8-14T07:39:07Z | |
dc.date.available | 2023-8-14T07:39:07Z | |
dc.date.issued | 2023 | |
dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=110523037 | |
dc.contributor.department | 通訊工程學系 | zh_TW |
DC.description | 國立中央大學 | zh_TW |
DC.description | National Central University | en_US |
dc.description.abstract | 隨著自動駕駛技術的發展,在車載資通訊環境下車輛裝置產生龐大且持續成倍增加的資料量,通訊網路系統不但需要提高資料傳輸效率,更要維持不同車載應用的服務品質。傳統的通訊網路系統主要在提供車輛裝置上行及下載的數據傳遞服務,當前自動駕駛車輛與ITS系統的迅速發展,車輛及其他鄰近車輛間的訊息交換與計算工作協作,遂發展出一新興的分散式資料傳輸模式,同時也擴大通訊網路系統在資料處理工作的負荷。近年來,車輛間資料傳輸之研究經常採用強化學習技術來解決任務卸載的問題,強化學習能夠在沒有事先知識的情況下學習並與環境互動,從而發展出一優化策略。
最近關於自動駕駛數據處理的研究文獻提出的許多方法在這些技術中,在計算卸載問題上並沒有考慮到車輛安全問題,在計算卸載量大幅增加的情況下,傳輸上恐發生過度延遲,或者,在高速行駛的狀況下強化學習的行為決策無法壓制在延遲門閥值之下。強化學習能夠在沒有先驗知識的情況下學習並與環境互動,從而發現最優策略。深度強化學習方法已經被探索出來,以實現最低的能源消耗或最小的延遲。
因此我們的研究提出了一套適用在車載資通訊環境下並具備低延遲的資料計算卸載機制。此機制建立在以5G-CV2X的行動網路環境下,本機制採納車輛本機端、多接取邊緣架構伺服器以及雲端伺服器的資料處理負荷等因素,此機制功能包括:透過深度強化學習來動態決定車輛資料計算的卸載和合作策略,使車輛能夠隨著車輛和道路情況動態調整卸載策略、透過計算卸載的方式,有效地分散資料處理負荷使車輛穩定性提高、實現車輛的低延遲卸載和高駕駛安全性。 | zh_TW |
dc.description.abstract | With the advancement of autonomous driving technology, the mass of in-vehicle devices generate a massive and continuously increasing amount of data in vehicular communication environments. Communication network systems not only need to improve data transmission efficiency, but also maintain the service quality of different vehicular applications.Traditional communication network systems mainly focus on providing data transmissions for in-vehicle devices′ uplink and download services. However, with the rapid development of autonomous driving vehicles and Intelligent Transportation Systems (ITS), a new emerging distributed data transmission model has emerged for facilitating message exchange and collaborative computing among vehicles and neighboring vehicles. This expansion places a heavier workload of data processing task on the communication network.
In the literature, recent research recent research on autonomous driving data processing has proposed various methods for computation offloading. However, many of these methods have not adequately considered vehicle safety. As computation offloading increases significantly, it may cases in excessive delays during data transmissions. In the high-speed driving, reinforcement learning-based decision-making cannot suppress the delays below the threshold.
To address these challenges, our study proposes a low-latency data computation offloading mechanism suitable for vehicular communication environments. This mechanism can operate in a 5G-CV2X mobile network environment and can be incorporated with several factors such as local vehicle capabilities, Multi-access Edge Computing (MEC) servers, and cloud servers for data processing loads. In addition, the functionality of this mechanism includes the dynamic determination of vehicle data computation offloading and the collaboration strategies using deep reinforcement learning. This mechanism enables vehicles to adjust offloading strategies dynamically based on vehicle and road conditions, effectively distributing data processing loads to enhance vehicle stability, achieve low-latency computation offloading, and ensure high driving safety. | en_US |
DC.subject | 計算卸載 | zh_TW |
DC.subject | 強化學習 | zh_TW |
DC.subject | 多接取邊緣架構 | zh_TW |
DC.subject | C-V2X | zh_TW |
DC.subject | Computing Offloading | en_US |
DC.subject | Reinforcement Learning | en_US |
DC.subject | Multi-access Edge Computing | en_US |
DC.subject | C-V2X | en_US |
DC.title | 在5G-CV2X 環境下利用Dueling Double DQN 最小化任務卸載時間之研究 | zh_TW |
dc.language.iso | zh-TW | zh-TW |
DC.title | On Optimizing Low-Latency Task Offloading with Dueling Double DQN in 5G-CV2X Environments | en_US |
DC.type | 博碩士論文 | zh_TW |
DC.type | thesis | en_US |
DC.publisher | National Central University | en_US |