近年來旅行時間預測成為智慧化運輸系統(Intelligent Transportation System，ITS)重要的議題，根據國際汽車組織(Organisation Internationale des Constructeurs d'Automobiles，OICA)的統計，全球的車輛不斷的增加，至西元2012年為止全球車輛已達到八千萬輛。而根據中華民國交通部(Ministry Of Transportation and Communications R.O.C.)統計指出，至西元2012年為止，台灣的國道使用率每年有五萬七千萬輛台小客車使用國道。增加的車輛數以及增加的道路使用率會間接地造成車輛壅塞，而在車輛壅塞的情況下駕駛者若因為搶快或者對於周遭環境不熟就很有可能造成車禍事故的發生。為了避免事故的發生以及保障用路人生命安全，精準的預測車流旅行時間可以讓用路者明確的瞭解道路的環境，進而避免自己陷入車輛壅塞的環境之中。傳統的旅行時間預測主要有兩種方法，第一種利用車內裝置定期的將車輛現在的資訊傳回中控中心，這種方法的缺點在於車載網路瞬息萬變，過度依賴全球定位系統(Global Position System，GPS)除了有可能面臨到通訊中斷而使得資料丟失，也有可能會因為GPS延遲使得中控中心取得不正確的資料；第二種是使用路側設施所獲取的車流資料，但是這種方法很容易因設施毀損以及通訊中斷導致中控中心無法取得正確的資料。為了精準的預測車流旅行時間，本論文提出了混合式之車流旅行時間預測方法HPAM (Hybrid Travel-Time Prediction Approach Based on Macroscopic and Microscopic Methodologies)，利用卡爾曼濾波考慮環境噪音的影響並且能夠快速的修正中控中心遺失的資料，並且利用支援向量機的方法考慮了車流資訊以及空氣汙染的影響，進而精準的預測車流旅行時間。利用本論所提出的HPAM機制，在高速公路環境下可以減少9.86%至54.40%的預測誤差，而在一般道路環境下可以減少9.75%至72.80%的預測誤差，因此本論文所提出的HPAM機制能夠有效的減少車流旅行時間預測誤差。 Recently, travel time prediction approaches have become an important issue in intelligent transportation systems. According to the Organisation Internationale des Constructeurs d’Automogiles , the amount of global vehicles is over than eight hundred millions in 2012. According to the Ministry Of Transportation and Communications R.O.C., the usage rate of Taiwan freeway is more than 5.7 billion vehicles in 2012. Increment of vehicles and freeway usage rate indirectly conducts traffic jams, and traffic jams might conducts accidents. In order to avoid the car accident and keep people safe, accurately predict travel time let users to know the road situation clearly. There are two methods of traditional travel time prediction. First, vehicles use on board unit or global position system to deliver their information to control center. This method has a critical disappoint like control center will lose data due to communication lose or transmit latency. Second, using vehicle detectors to retrieve vehicle information and then transmit to control center. This method also faced same problem like first method. In order to predict travel time more accurately, this thesis propose a Hybrid Travel-Time Prediction Approach Based on Macroscopic and Microscopic Methodologies (HPAM) , using Kalman Filter to recover the loss data quickly and then using Support Vector Machine to predict travel time. In this thesis, HPAM can reduce 9.86% to 54.40% prediction error in freeway situation, and reduce 9.75% to 72.80% prediction error in urban situation. The mechanism HPAM can accurately predict travel time.
