應用貝氏方法針對肇事碰撞型態建立號誌化交叉口機車交通事故因子分析模式

Applying Bayesian Models to Analyze Motorcycle Crashes at Signalized Intersections by Collision Types

顏愉(Yu Yen)

指導教授 ： 許添本

國立臺灣大學/工學院/土木工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.01887

摘要

本研究以分析號誌化交叉路口影響機車肇率之因子為目標，依據常見之號誌化交叉路口碰撞型態:同向右轉側撞、左轉穿越側撞、鄰向交叉撞與同向直行擦撞，分別建立貝氏統計模型，分析找出肇事影響因子。
分析資料包含高雄市24個與新北市8個，共32個分析交叉口。本研究建構了層級Poisson-gamma與Poisson-lognormal模式，針對四個肇事碰撞型態進行分析。以模式比較指標DIC比較結果顯示，對於四個肇事碰撞型態之統計分析，均以層級Poisson-gamma模式表現較優。
肇事影響因子分析結果顯示同向右轉側撞(機車-汽車)方面，角度1(與90度之差)、本向有無設置快慢分隔島、本向進入路口總車道數、本向右轉與直右車道數、本向右轉汽車與重車交通量對同向右轉側撞(機車-汽車)率有正向影響，而本向機車可行駛之車道數的設置對於同向右轉側撞(機車-汽車)則有減少的趨勢。而對於左轉穿越側撞(機車－汽車)之肇事率，本向快車道數、對向是否設置左轉專用時相、黃燈秒數、本向直行機車交通量等均有正向影響。鄰向直行交叉撞(全－機車)分析結果顯示，包含角度1(與90度之差)與本向有無設置快慢分隔島對肇事率有反向的影響。而本向機車兩段式左轉、本向理論與現況全紅秒數差、本向直進與鄰向直進機車交通量則對鄰向直行交叉撞(全－機車)肇事率有正向之影響。而同向直行擦撞(機車－全)分析結果顯示，是否為正交路口、本向進入路口總車道數、本向路口分支機車百分比對同向直行擦撞(機車－全)肇事率有正向影響，而本向有無設置快慢分隔島則有反向影響。

關鍵字

號誌化交叉路口 ； 交通安全 ； 機車 ； 肇事分析 ； 碰撞型態 ； 貝氏統計

並列摘要

Since the motorcycle is the most common type of vehicle used in Taiwan, it accounts for the largest proportion of all of Taiwan’s vehicles. Consequently, it is important to accurately understand the frequency of motorcycle crashs. However, the mixed traffic characterists (including motorcycles and cars) on Taiwan’s roads increses the difficulty of modeling such crashes. The objective of this study is to identify the impact factors of crashes at signalized intersections. This study applied Bayesian models in an effort to analyze motorcycle crashes according to the following common collision types that occur at signal-controlled intersection: through with right turn, through with opposing left turn, right angle, and sideswipe from the same direction. A total of 128 approaches to 32 four-leg intersections in Taiwan were used for this analysis. Two Bayesian models, the Hierarchical Poisson-gamma and Poisson-lognormal models, were explored. Compared to DIC, the Hierarchical Poisson-gamma model was found to be better than the Poisson-lognormal model for all four collision types. The results of this study show the relationship between the crash frequency of motorcycles and the intersection characteristics that occur in mixed-traffic flows, which is representative of the unique road designs and traffic control methods common in Taiwan (including the presence of express / slow traffic dividers, the number of motorcycle permit lanes, and the number of fast lanes). The results indicate that different collision types can be attributed to different accident factors. The results also show that an accident factor analysis performed according to collision type can be useful in describing the targeted causes. It is expected that the results of this study will help to develop more effective corresponding safety countermeasures in areas of Southeast Asia faced with such mixed traffic flows, such as Taiwan, China, and Viennam.

並列關鍵字

signalized intersection ； traffic safety ； motorcycle ； accident analysis ； collision type ； Bayesian inference

參考文獻

8. 林沛婕, 號誌化 T 字路口機車左轉管制設置準則綜合評估之研究. 臺灣大學土木工程學研究所學位論文, 2013: p. 1-148.

1. Ma, J., K.M. Kockelman, and P. Damien, A multivariate Poisson-lognormal regression model for prediction of crash counts by severity, using Bayesian methods. Accid Anal Prev, 2008. 40(3): p. 964-75.

2. Dong, C., et al., Examining signalized intersection crash frequency using multivariate zero-inflated Poisson regression. Safety Science, 2014. 70: p. 63-69.

3. Pai, C.-W. and W. Saleh, Modelling motorcyclist injury severity by various crash types at T-junctions in the UK. Safety Science, 2008. 46(8): p. 1234-1247.

4. Wang, X. and M. Abdel-Aty, Analysis of left-turn crash injury severity by conflicting pattern using partial proportional odds models. Accid Anal Prev, 2008. 40(5): p. 1674-82.