人行道行人流量之推估與應用

Sidewalk Pedestrian Flow Estimation and Applications

本研究旨在建立人行道行人流量推估模式，並用以評估人行道服務狀況及排定改善優先順序。本研究利用迴歸分析及倒傳遞類神經網路模式，建立行人流量與27個解釋變數之關係，包括人行道幾何條件、社經人口變數、土地使用變數、公共運輸變數及POI (place of interest)。平假日行人流量推估結果顯示，分別計有15、10個顯著變數，配適度佳(adjusted-R2=0.65及0.73、驗證MAPE(mean absolute percentage error) =7.18%、9.60%)。倒傳遞類神經網路模式則驗證MAPE= 9.67%、8.49%。兩種模式都獲得相當精準之預測結果。其中，影響平假日行人流量高低的關鍵變數為土地使用及人口結構。本研究亦利用分析階層程序法建立人行道服務狀況評估，最後搭配各人行道推估行人流量高低，以二維圖方式展現臺北市人行道分佈情形，有助於指認其改善優先順序。

This study aims to develop pedestrian flow estimation models of sidewalks so as to evaluate the level of service of the sidewalks and identify their improvement priorities. The regression model and back propagation neural network (BPN) model are used to estimate pedestrian flow based on a total of 27 explanatory variables including sidewalk geometrics and facilities, social economic and demographics, land use variables, public transportation, and POI (place of interest). The estimated double-log regression model for weekday and weekend pedestrian flows shows 15 and 10 variables are significantly tested with adjusted-R2 of 0.65 and 0.73 and MAPE (mean absolute percentage error) 7.18% and 9.60%, respectively. The tuned BPN model shows the MAPE of 9.67% (weekday) and 8.49%. Both regression and BPN models perform very satisfactorily. To show the applicability of the estimated models, this study further develops an evaluation framework to measure the level-of-service of sidewalks by using analytic hierarchy process (AHP). A case study of sidewalks in Taipei city is then conducted. The distribution of these sidewalks on a two-dimension figure (pedestrian flow vs. LOS scores) is helpful to prioritize their improvement.