高速公路車輛的行駛速率較快，且道路系統較為封閉，一旦事件發生，對駕駛人生命財產的威脅及車流延滯的影響將遠較一般平面道路嚴重。因此，發展迅速且正確的高速公路自動事件偵測演算法，不但能提供交通管理者事件發生時的相關資訊，讓交通管理者能儘早將事件排除，以減少事件對高速公路車流的衝擊外，也可讓事件造成的損失減到最小。 台灣地區自2004年起開始建置電子收費系統，屆時將會利用自動車輛辨識技術來收取通行費用。而自動車輛辨識技術除了可用於收費外，同時也可取得車流參數以應用於高速公路自動事件偵測中。由於國內應用自動車輛辨識技術於高速公路事件偵測尚處於起步階段，因此本研究發展一套(1)利用自動車輛辨識技術獲得交通參數資料，(2)利用自動車輛辨識技術所獲得的交通參數-旅行時間及平均速率，並結合統計方法中的統計分配配適與無母數檢定，來判斷事件發生與否的高速公路自動事件偵測演算法。在績效分析部分，由於國內尚未建置自動車輛辨識技術相關設施，因此本研究利用車流模擬的方法以及三項績效評估因子：偵測率、平均偵測時間以及誤報率等來評估自動事件偵測演算法的績效。 經由一系列的評估分析，整理得本研究所提出的事件偵測演算法其偵測率為97%，平均偵測時間為8.58分，誤報率為0.576%。相較於文獻中的事件偵測演算法，本研究所提出的事件偵測演算法具有高偵測率、低誤報率的優異表現，在平均偵測時間方面也屬可接受的表現。因此本研究的結果確可提供相關單位研究或建置系統參考用。 Incidents on freeways generally cause tremendous calamities than those on other types of road systems due to its closed system and higher speed limit. These calamities always include serious traffic delay and the loss of lives and financial affairs. Developing rapid and correct incident detection algorithms for freeway automatic incident detection can minimize damages from incidents, because it provides incident-related information for the traffic operators who can exclude the incident quickly and reduce the impact of freeway traffic flow efficiently. In 2004, Taiwan is going to build up an electronic toll collection system applying automatic vehicle identification. Except for collecting tolls, automatic vehicle identification can gather traffic parameters for freeway automatic incident detection. Worldwide, there has been little literature published about automatic vehicle identification on automatic incident detection. In Taiwan research in that subject has just started in the recent years. This research develops an automatic incident detection algorithm which combines: (1) automatic vehicle identification to obtain traffic parameters, and, (2) statistic methods to judge the incident happened or not. Because of automatic vehicle identification facility not being constructed yet, this research verifies the performance of the incident detection algorithm by utilizing traffic simulation. After series analysis of the incident algorithm, it is concluded that the detection rate is 97%, mean detective time is 8.58 min, and false alarm rate is 0.576%. Comparing with the algorithms in literature, it is shown that performance of this research is greater than those in detection rate and false alarm rate. And the performance in mean detective time is also acceptable. Therefore, the algorithm developed by this research can be provided to other researches and relative associations for reference.
