建立風險決策模式於路段機車空間管制

Establishment of Risk Decision Model for Motorcycle Riding Space Control

張哲寧(Che-Ning Chang)

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(2016年)

https://doi.org/10.6342/NTU201601110

摘要

根據交通部統計，2015年台灣機車占有率高達64%，然而三車道以上路段普遍劃設「禁行機車」，為了保障機車合理行駛空間，政府自民國98年起採個案檢討例外開放原則陸續取消外側第三車道禁行機車，但是目前並無對於取消外側第三車道禁行機車事前事後影響較為詳細之分析，因此本研究以交通安全的角度探討該措施對於台灣交通的影響。
首先蒐集目前台北市自民國100年至民國105年取消外側第三車道禁行機車之肇事資料，並以完全貝氏法建立兩個貝氏統計模型Hierarchical Poisson-Gamma model和Hierarchical Poisson-Lognormal model，可求得且分析各車道肇事影響因子，並以DIC值比較模型適合度，結果以Hierarchical Poisson-Gamma model表現較優。
接著計算所蒐集路段事前與事後之風險值，並應用於決策樹(CART)，定義比較事前事後之風險為惡化或改善，以此作為決策樹之目標變數，道路幾何特性與交通量特性則作為自變數，可以得到七條規則，並以隨機森林找出影響決策之重要變數；並比較所有因子、貝氏因子與隨機森林因子建構出之決策樹，以錯誤矩陣驗證經過篩選因子後貝氏因子與隨機森林因子之決策樹整體誤差僅19%，可作為未來政府取消外側第三車道禁行機車之參考。

關鍵字

禁行機車 ； 事前事後分析 ； 肇事因子分析 ； 貝氏統計模型 ； 資料探勘

並列摘要

According to the statistics of the ministry of transportation and communications, the sharing rate of the motorcycle is 64% in 2015. However, in generally, the three-lane section’s inside lane is set to be “Forbidding-Motorcycles”. In order to protect the riding space of motorcycles, the government gradually cancelled the forbidding-motorcycles of the third lane based on the local conditions. Nevertheless, there is no detailed analysis of before and after of cancel the forbidding-motorcycles of the third lane. In this case, this study aims to find the influence of motorcycle policy in the safety of traffic measure.
The study collects the accident data of forbidding-motorcycles of the third lane from 2011 to 2016 in Taipei. Using the Full Bayesian Method to build two Bayesian Statistical Models: Hierarchical Poisson-Gamma model and Hierarchical Poisson-Lognormal model. With the above models, the accident factors of each lane can be obtained. The DIC value can be applied to define the fitness degree of the model. The result shows that Hierarchical Poisson-Gamma model has better performance.
Finally, analyzing the traffic characteristic and accident data which before and after the forbidding-motorcycles, and computing the risk of each road. This result is used in the CART to fine the difference of traffic situation because of the motorcycle policy. After setting the result to be the decision variable of CART, and the road geometric characteristics and traffic volume as the independent variable, the seven rules are gained. The random forests can be applied to find the key variables. Comparison of decision trees which construct by all factors, Bayesian factors and random forests factors. The results show the error is 19% of decision trees which construct by all factors, Bayesian factors and random forests factors, which is comparatively low. This CART can be the effecttive reference for the government to decide whether the forbidding-motorcycles policy of the third lane cancelled or not.

並列關鍵字

Forbidding-Motorcycles Policy ； Before and After Influence Analysis ； Accidental Factor Analysis ； Bayesian statistical model ； Data Mining

參考文獻

[5] 林沛婕(2013)，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

[7] 顏愉(2015)，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. 臺灣大學土木工程學研究所學位論文

[16] 蕭唯倫(2015)，應用資料探勘技術針對肇事碰撞型態建立路口分支風險和事因子模型. 臺灣大學土木工程學研究所學位論文

[21] O’Donnell and Connor, (1996). Predicting the severity using models of ordered multiple choice. Accident Analysis and Prevention, 28 (6) (1996), pp. 739–753

[22] Al-Ghamdi (2002). Using logistic regression to estimate the influence of accident factors on accident severity, Accident Analysis & Prevention 34(6):729-41