請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/74402
標題: | 基於類神經網路方法之最大化幹道續進動態號誌控制策略 A Dynamic Signal Control Strategy to Maximize the Progression on an Arterial Road: A Neural Network Approach |
作者: | Zhi -Xun Xu 許智勛 |
指導教授: | 許聿廷(Yu-Ting Hsu) |
關鍵字: | 動態號誌控制,類神經網路模型,幹道續進,交通需求預測,交通管理, Dynamic signal control,Neural network model,Arterial road progression,Traffic demand prediction,Traffic management, |
出版年 : | 2019 |
學位: | 碩士 |
摘要: | 在每日上下班的尖峰時段,市區幹道總是充斥著壅塞造成大量的時間成本與燃油消耗,而經過周詳設計的號誌控制則可以有效的緩解壅塞。近年來,由於電腦計算資源的改善,透過即時運算來優化號誌控制,進而實現動態交通控制的相關研究如火如荼的進行。然而,在優化號誌的過程中,建構一個符合當地交通特性的代表模型,需要校估大量的參數,這過程是相當複雜且繁瑣的。另外,針對每個不同的交通情境,需在有限時間內求解有效率的交通控制策略,這對於實現動態號誌控制又是另一個棘手的問題。目前,在幹道層級的動態號誌控制的整體決策上,所需的計算量仍是相當大的,導致需要較多的運算時間。因此,在實務中,透過蒐集即時交通資訊,進而調整相對應的號誌控制策略仍然是相當富有挑戰性。有鑑於上述提及實現動態號誌控制的兩個主要限制,本研究利用已蒐集的大量歷史車流資料與交通號誌控制參數,在巨觀的觀點下,應用類神經網路,自動且有效率的架構出符合當地交通特性的代表模型與校估模型中的大量參數。更重要的是,本研究提出一套分解流程,將類神經網路所辨識且紀錄於其中的樣態逐一分解,用以架構號誌最佳化模式。此方法能顯著的改善求解號誌控制策略所需的時間,進而實現動態號誌控制。此外,一個預測短期交通流量的類神經網路模型也被建立,用以更新即時車流資訊,制定適當的動態號誌控制策略。本文的案例研究為連結國道一號高速公路的幹道。由於國道車流與幹道車流分屬不同道路層級,兩股巨大車流不斷匯聚導致頻繁的煞停,且車流量的即時變動性高,都可能是造成壅塞的原因。動態號誌控制依據即時車流資訊,及時且不斷的調整策略,以促進幹道車流有效續進,使匯聚後的大量車流能盡速紓解,這說明了其具備足夠的效益。透過案例研究的結果,說明本研究所提出的動態號誌控制策略,具備可行性與維持幹道續進的穩健性,也進而凸顯出動態號誌控制所具備的彈性調整優勢。整體來說,本研究所提出的動態號誌控制策略能迅速的挖掘巨量的歷史資料中所蘊含的交通樣態資訊,用以架構交通代表模型,並即時優化號誌策略。期望能提供實務上,一套自動且有效率的動態號誌控制策略之決策程序,以促進交通管理的效率。此外,本研究方法亦可進一步延伸,由類神經網路自行發展適應性號誌控制邏輯,具備研究潛力。 In peak hours, congestion on urban roads leads to considerable extra time cost and fuel consumption. A well-designed signal control strategy can efficiently relieve congestion. Due to the improvement of computing resource, realizing dynamic signal control strategy becomes possible. However, estimating several traffic parameters to construct a model fitting real traffic conditions can be intricate, and solving the efficient signal control strategy in finite time also introduces another methodological difficulty. Currently, the solution time for determining a signal control strategy at an arterial road level can be still long. Hence, the concept of adjusting the control strategy based on the real-time traffic state is challenging in practice. This study applies the ANN model to construct a model fitting real traffic conditions and estimate a considerable number of parameters efficiently and automatically. Most importantly, a decomposition method is proposed in this research to decompose the patterns recorded in the ANN model and construct a signal optimization model, thereby improving the computational time of a signal control strategy significantly. A traffic demand prediction model is also developed to complete a dynamic signal control procedure. The arterial road connected with the Freeway No.1 of Taiwan is selected for the case study, which is an arterial road of great traffic volumes and complex flow merging and weaving, causing frequent occurrence of serious congestion. The results of the case study show the feasibility and robustness of the proposed approach to promote progression and highlight the advantage of adjusting the control strategy in a more dynamic manner. The proposed dynamic signal control strategy is expected to maximize the use of the historical traffic data and provide a credible and dynamic adjustment signal control module to relieve or even prevent congestion. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/74402 |
DOI: | 10.6342/NTU201902935 |
全文授權: | 有償授權 |
顯示於系所單位: | 土木工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-108-1.pdf 目前未授權公開取用 | 2.51 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。