汽車及機車於市區道路混合車流之行為模式

Abstract

國內市區道路混合車流以汽車、機車和公車為主要車種，台灣地區道路中的機車數量龐大，微觀模擬軟體無法完整呈現國內機車在混合車流中之移動行為，因此造成模擬預測準確性下降。本研究針對汽機車進行路段中的行為模式分析，過去微觀混合車流研究受限於原始資料收集不易，但是近年來隨著影片拍攝設備不斷進步，配合數位化的資料收集軟體也開始發展，本研究建立台北地區的混合車流資料庫，並且以自動化的方式進行鄰近車輛與變換車道判斷，分析國內市區道路為主體的混合車流駕駛特性與行為模式，可以協助改善市區道路模擬預測與號誌控制效能，使整體市區道路路網更加順暢。 本研究針對小汽車進行變換車道模式之研究，以多項羅吉特模式建立小汽車的變換車道決策模式，將直行、向左偏移、向右偏移作為主要的三種決策，主要考量的變數為相對距離、相對速度、鄰車車種，本研究將本車速度也納入考量，並且將前車與(左前車和右前車)分開校估。完成利用國內原始資料校估的小汽車變換車道模式。並且以判中率來測試小樣本情況下的判中情形，整體判中率結果為70%；如果將樣本切割測試也有63%的判中率。 本研究對國內市區道路的機車駕駛行為特性進行分析，觀察台灣機車的超車與鑽車等特殊行為。為了瞭解在市區道路中的不同行為模式，將機車駕駛的鄰車以鄰近區塊的方式來擷取不同周邊車輛資訊，用客觀的方式將不同的駕駛者類型進行分類，使用K-Means分群法將機車駕駛人分成不同冒險程度的類型，利用兩種分群參數來將不同號誌週期內的機車駕駛分群。型態A表示較激進的駕駛者；型態C是較緩慢保守的駕駛者；型態B駕駛者較多，為介於兩者之間的機車群。本研究將資料依照前方區塊影響情境 (Scenario 1,2,3)分開討論，可以判斷不同對應車種對機車駕駛的影響，最後將所用來分群的資料點使用規則來判斷每位駕駛者的駕駛型態。以3種情境和3種駕駛類型可以形成一3×3的分類矩陣，矩陣內會對應不同的觀察機車移動行為，可以針對不同組合進行模式分析。

The urban arterials in Taiwan are characteristics with a mixed traffic flow composed of cars, motorcycles, and buses. As the amount of motorcycles traveling on the road in mixed traffic is huge, the traffic simulation software developed in foreign countries does not have the capability to model the characteristics of traffic in Taiwan. As the simulation of motorcycles is unrealistic, the modeling results are usually inaccurate. The aim of this study is to analyze the driving characteristics and behavior of cars and motorcycles in mixed traffic for urban arterials, which may shed some lights on the modeling and signal control design of urban network. A case study is done based on a selected site location in Taipei city. We collect traffic data by taking video of traffic movements, digitalize it into trajectories and movements of vehicles, and build a database for analysis. We formulate lane-changing of cars and motorcycles for urban arterials as a discrete choice model using multinomial logit model, considering the three decisions of a driver as moving to the left, going straight and moving to the right. The explanatory variables include relative distance, relative velocity, neighboring vehicle mode and the velocity of vehicle etc. We also study the driving behavior of motorcycles, such as overtaking and filtering. Based on the dataset, it is observed that different motorcycle drivers may behave differently in mixed traffic, and there is a need to categorize the drivers for further analysis. We propose a concept of neighboring blocks to capture the nearby vehicles of a subject vehicle. With factors from neighboring blocks, a K-Means clustering approach is proposed to categorize the data and thus motorcycle drivers into three types (i.e. aggressive, normal, and conservative). The clustering is based on three scenarios that a subject vehicle is facing in mixed traffic (i.e. neighboring with motorcycles, cars, and buses). Therefore, the collected dataset can be categorized into a three by three matrix for the relationship of driver types and scenarios, and the results can help to realize the driving behavior of motorcycles.