綜觀交通事件相關研究，國內外文獻於事件偵測與預防、事件發生後交通狀況分析與預測以及事件管理等類型皆有許多探討，但於事件發生後，對時間與空間衝擊擴散影響情形之研究，則未如其他課題受到重視。但完善之事故管理系統，除運用各種事故偵測或通報系統外，評估並掌握事件造成交通衝擊，以利交通管理權責單位採取即時且有效的管理措施是必要的，因此，更需要一套方法論以預測事件對整體範圍的衝擊影響。 本研究之模式構建分為兩部分：第一部份為受事件影響路段數遞增與遞減兩函數，其中以事件降低容量嚴重度與發生事件路段之擁擠指標為模式參數，以此二函數求得受事件影響路段數。第二部份以一路段指派原則，將求得路段數指派入路網，當所有路段數皆指派完畢，即可得到事件發生後對上游造成之影響區段範圍。 本研究之模式構建分為三階段：首先將高速公路路網劃分為等長路段，蒐集事件發生前後各路段速率於不同時間之差異，並依服務水準之變化判斷路段是否受事件影響，進而得到事件發生後，受影響路段數之遞增與遞減情形；第二部份以受影響路段數目與發生事件路段之擁擠指標，為事件影響區段範圍模式之輸入項目進行構建；最後模式完成後，藉由輸入事件嚴重度與發生事件路段之旅行時間變化等資料，得到受影響路段數進行整體範圍預測。 由於我國並無搜集事件發生後車流狀況之相關資料，故以模擬工具PARAMICS構建路網，並彙整國道所佈設偵測器之資料進行模擬模式參數校估，以建立擬真之交通環境；此外分析由國道警察提供之歷史事件資料，找出易肇事路段作為研究對象，並以不同流量環境、事件造成容量降低程度與發生位置等因素，設計數種事件情境構建模式。 最後本研究依據不同情境組合共彙整36種模式，其預測能力以MAPE指標進行驗證，結果皆屬良好與合理之範圍。其中發現擁擠擴散情形模式化結果較優於擁擠之消散且較嚴重之事件情境模式預測準確度較高。綜合以上，本研究針對高速公路事件發生所造成影響，提出一套預測之方法，以連鎖、交互影響之概念描述事件造成之擁擠情形，利用廣義統計模式之技術，配合經真實車流資料校估之模擬環境所得資料構建模式，並可藉由發生事件情境與部份路段資訊，透過模式運作得到路網受事件影響範圍。本研究成果可應用於ITS事故管理領域。 Traffic incidents occur in variety of forms in road network and subsequently cause traffic congestion and travel time delays. In such conditions travel times may be increased not only on the incident link, but also on the links which are the upstream links of the incident location. These upstream links can therefore be identified as links affected by the incidents, namely, the “Affected Upstream Sections” and the prediction of how many links being affected is a vital issue to the development of advanced incident management systems in modern Traffic Management System. Unfortunately this particular issue has been long ignored comparing to the other incident detection issues.
In this thesis, a new modeling approach originally proposed by Hounsell and Ishtiaq (1997) has been considered in which an “incident data base” was compiled using a simulation tool applied to a range of traffic and incident scenarios. A set of parameters was defined and the associated effects of these parameters were analyzed. Generalized statistical models were then developed to predict the number of links which would be affected by an incident of given characteristics. Two particular models work hand-in-hand for prediction of the number of effected links during the time period of formation (M1) and dissipation (M2) of the congestion caused by incidents respectively. Models performance was evaluated by statistically analyzing the prediction errors.
The data base for the statistical modeling was compiled by using Paramics, a microscopic traffic simulation program, applied to a variety of traffic and incident scenarios. This simulation was calibrated carefully utilizing its feature for detailed road section geometric configuration and driving behavior using true real world flow data before its application in this study. Two freeway sections with different characteristics were selected in this thesis.
The results show that the proposed modes have demonstrated a reasonable predictive quality where the M1 models performed better than the M2 in general. In addition, models specified for the scenarios with more severe incidents have exhibited better performance.
