中山高速公路肇事分析模式之構建－－一般化線性模式之應用

Development of Accident Analysis Models for Sun Yat-Sen National Freeway : Application of Generalized Linear Models

本文針對中山高速公路民國80年至84年之列管肇事資料進行肇事變動趨勢、肇事分配及肇事影響因素分析，並採用一般化線性模式(GLM)中之卡瓦松迴歸及負二項迴歸技術'分別構建三種路段(主線、交流道及收費站)之肇事分析模式。研究結果顯示三種路段之肇事分析模式有所差異，除模式解釋變數不盡相同外，亦各有不同之適用模式:收費站路段適用卡瓦松模式，而主線及交流道路段則以負二項模式較佳，顯示主線及交流道路段之肇事發生存在過度離散(即肇事發生次數之變異數大於平均數)情形。本文發現下坡路段、戰備跑道、交流道街突點數及速限等均對中山高速公路肇事次數有顯著影響;為改善肇事情形，建議在主線下坡路段普設超速測速照相、收費站宜避免設置於下坡路段終點、於戰備跑道中央分隔處連續佈設可移動式之紐澤西護欄塊、交流道設計應注意分散車流衝突，點、速限低於100kph之主線路段研究提高為100kph之可行性。

This paper analyzes the characteristics of accidents for the Sun Yat-Sen National Freeway. Fatal and severe injury data from January 1991 to December 1995 are used to construct three accident analysis models for different road sections via poisson and negative binomial regression techniques, two special methods of generalized linear models. The results show that the variables involved in each model as well as the appropriate regression techniques of different models are not all the same. Poisson regression model can well describe the toll station sections while negative binomial regression model appropriately depicts the mainline and interchange sections, representing a high variance-mean ratio of accident frequency. It is found that steep downgrade,warlike preparation runways, number of conf1icts on interchange sections, and upper speed limit are major factors inf1uencing the accident frequency. Several countermeasures to reduce the accidents on this freeway are also proposed.