以結構化社會經濟因素探討旅次發生-類神經網路與多元迴歸之比較

Abstract

旅次發生之推估在程序性運輸需求分析中是一相當重要的步驟。過去之 研究均以社會經濟因素推估旅次發生，然而，大部份都是以一地區或一個 家戶之社會經濟變數總體值為解釋變數，甚少考慮社會經濟變數總體值之 組成結構變化對旅次發生之影響。由文獻回顧中可知，隨著經濟的發展以 及社會之變遷，使得社會經濟變數不僅在量方面有所改變，在組成結構方 面亦有所變動。例如:人口年齡結構老化、女性就業人口之增加、都是三 級產業之增加等等均與過去有所不同。除此之外，由於不同目的之旅次發 生特性均有所不同，故本研究僅針對數量最多且發生於尖峰時段之家-工 作旅次進行研究。因此，本研究將社會經濟變數之結構性變化納入模式中 ，探討結構化之社會經濟變數對家-工作旅次發生之影響。 在實務方面 最常採用迴歸分析法推估家-工作旅次之發生，然而，迴歸分析法之應用 須符合許多基本之假設，而實際之資料很難完全符合迴歸分析法中之諸多 假設。近年來有部份學者以類神經網路中之倒傳遞網路法進行旅次產生之 推估，通常都能得到較迴歸分析準確之結果，是以本研究亦嘗試以類神經 網路法分析結構化社會經濟變數對家-工作旅次之影響，並比較其與多元 迴歸法之分析結果。 由實證分析可知，以結構化之社會經濟變數推估 家-工作旅次產生，除了能更深入了解不同結構之社會經濟變數對家-工作 旅次產生之影響，而且在部份之結構化模式中還能提高旅次產生推估之經 確度。另外，在家-工作旅次之推估方法方面，多元迴歸分析法與類神經 網路分析法亦各有適用之對象，在交通分區旅次產生總量模式方面，以多 元迴歸分析家-工作旅次之結果較佳，且能解釋模式之意涵，而在旅次產 生率模式方面，尤其是交通分區旅次產生率模式，則以類神經網路法推估 家-工作旅次可得較佳之結果。 Trip generation is a major step in four-step transportation demand analysis.In the past,the socioeconomic factors considered in trip generation model were income,population,employment,and number of vehicles in each traffic zone or household.However,the massive changes in the socioeconomic structure such as the rise in elder people,the increase in the number of women at work and the growth in the service sector,etc.,has caused the change of trip generation.Therefore,this study tries to establish the trip generation model with structural socioeconomic fators for the largestand the most important home-based work trips. Most studies constructed home-based work trip generation model by usingregression analysis which must follow many assumptions. Recently an alternativemethod,artificial neural network,has been proposed and examined for trip production.And the results were better than that analyzed by using regression analysis.So this study uses artificial neural network and regression methods to estimatehome-based work trips production and compare their results. The empirical study shows that using structural socioeconomic fators inregression models to estimate home-based work trips would get more informationfor trip production.In addition,the estimation accuracy in some models is higherthan that in traditional regression models.The model comparison shows that usingartificial neural network models is more appropriate for estimating the trip rate than the regression models.But the regression models are more appropriatefor estimating the zonal trip totals than the artificial neural network models.