建立巨微觀機車混合車流模式

Abstract

按照詳細程度來分，車流模式可以分為：巨觀(Macroscopic)、中觀(Mesoscopic)與微觀(Microscopic)三種模式，分別應用時機關係到準確性與效率之間的權衡。巨觀車流模式在於求解大量車流行為之描述，優點在於處理時間效率高但準確性較為觀車流模式低；而微觀車流模式則針對於車輛與車輛間之刺激與反應，優點在於資料之詳細程度。為了呈現於長廊路段單純之車流行為與接近路口的複雜的車隊等候，分別利用巨觀模式描述單純的長廊路段與利用微觀模式描述鄰近路口路段可確保同時具有準確性與時效性。然而，如何求得最佳之巨微觀中間介面位置與如何設計界面資訊轉換機制即是巨微觀模式的核心問題。再者，為了解釋亞洲市區道路內主要小客車與機車混合車流，建立巨微觀混合車流模式有其必要性。 基於上述觀點，本研究嘗試建立一巨微觀車流模式以描述台灣混合車流。本研究所提出之巨微觀混合車流模式的創新思維有三點：1. 結合兩種模式的優點，並可正確流暢傳遞車流之資訊；2. 考慮機車插空隙的性質，並透過車種比例決定車種的擁擠密度；3. 加入多車道的模式，並考慮車道之間車輛的互動與變換車道。在模式驗證上，本研究以兩架攝影機自高樓拍攝一段長度為300公尺之三車道都市公路，並透過每秒擷取畫面記錄每輛小客車與機車之位置軌跡於長廊路段與鄰近路口路段。 本研究於不同介面設置位置比較所提出之模式的表現差異。準確性之評估準則係利用對稱平均絕對誤差百分比(SMAPE)比較實際資料與模擬流量每秒之結果於不同車道之小客車與機車。如同預期，介面越靠近路口，效率性越高但準確性則降低。然而，效率性改善與準確性降低的程度皆隨著介面越靠近路口而降低。小客車車流與機車車流均有合理之模擬結果。研究結果發現模擬機車車流之誤差率較小客車流量大，可能造成原因為受到轉向車流和路邊停車之影響。此外，長廊路段之誤差大於鄰近路口路段之誤差，因為車流於鄰近路口路段受到號誌影響。內車道之誤差較其他車到小，因為快車道禁止機車行駛。整體而言，本研究所提出之巨微觀模式表現是可靠的。

In light of the level of details, three categories of traffic flow models: macroscopic, mesoscopic, and microscopic, are independently adopted for the trade-off between simulation accuracy and efficiency. The macroscopic models accounting for the behaviors of a group of vehicles are obviously more efficient but less accurate than the microscopic models which replicate the movements of individual vehicles. However, to replicate the traffic movement behaviors along a corridor containing segments with simple traffic behaviors and intersections with complex queuing, weaving and turning behaviors, it is rationale to use of macroscopic models to simulate traffic behaviors moving along segments and use of microscopic models to replicate traffic behaviors approaching intersections for compromising simulation accuracy and efficiency. However, how to determine the optimal location of the interface of macro- and microscopic models and how to design the interfacing mechanism to convert macroscopic flows into microscopic flows are at the heart of the integrated models, namely, the macro-micro models. Furthermore, to acknowledge the prevailing mixed traffics of cars and motorcycles on Asian urban streets, mixed macroscopic and microscopic models are considered. Based on these, this study attempts to develop a macro-micro model which can account for the mixed traffic condition in Taiwan. The novelties of the proposed macro-micro model are threefold: 1) Ability to convert upstream macroscopic traffic flow into downstream microscopic flow. 2) Consideration of the lateral drifts and transverse crossings of motorcycles by determining the jam density and free flow speed according to car-motorcycle density ratio. 3) Consideration of lane changing behaviors at the roads with more than three lanes. To calibrate and validate the proposed models, a video-taking of a three-lane urban street was conducted. Trajectories of all cars and motorcycles within the study segment (300m in length) and intersection were then frame by frame traced and recorded. The performances of the proposed model under various locations of the interface are compared. Accuracy index is measured by the SMAPE values between real and simulated second-flows of cars and motorcycles at different lanes and distances to the intersection. Efficiency index is of course measured by the simulation times. As expected, the closer to the intersection of the interface is, the higher efficiency but the lower accuracy it has. However, the improvement in efficiency and deterioration in accuracy is diminishing as the distance to the intersection decreases. Both flow of cars and motorcycles are well simulated. It is also found that the error rates of simulated motorcycle flows at different lanes are higher than those of simulated car flows. One of the reasons for that may be the influence of turning flow and parallel illegal parking. Furthermore, the error rates at the mid-road section are higher than at the intersection because the traffic flows are restricted by the signal control. The error rates of traffic flows in inner lanes are lower than those of outer lanes because of the prohibition of motorcycles in using the inner lane. Overall, the performance of the proposed model is promising.