Scholarlink

<p>Travel times for many individual origin–destination pairs can be estimated without relying on detailed transport simulations by using commercial mapping services or by combining open-source routing engines with open street network data. While the former are more accurate, the latter are more scalable and economical. Neither option typically accounts for the time it takes to get to and unlock a vehicle or find parking. Here, we present a data-driven and scalable approach to obtain accurate, representative door-to-door travel time estimates for walking, cycling, e-bikes, and driving using open network data. To do so, we first use the open source routing machine (OSRM) to produce baseline estimates, then update these estimates based on characteristics of the built environment along the route and near the origin and destination of each trip. To apply this approach, we harness multiple travel survey datasets from Switzerland and evaluate the predictive performance of two statistical models: a linear regression and the XGBoost machine learning algorithm. We find that both models are capable of producing substantially more realistic travel time estimates than the baseline routing engine alone, achieving correlations with the Google Maps API of over 98%. Key predictors include intersection count and density along the route. Topography, critical for active modes, can effectively be incorporated using a constant time penalty per meter of elevation gained and lost. XGBoost is 5–20% more accurate than the linear model, depending on the mode. On the other hand, the coefficients from the linear model can be used to update the original routing engine to produce representative door-to-door travel time estimates without subsequent statistical processing. Finally, we highlight the importance of such estimates: Without adjustments, pedestrians and even bicycles are estimated to be faster than cars less than 3% of the time. With our adjustments, during peak hours and in or nearby cities, bicycles are faster than cars for 60% of all trip legs, and e-bikes for up to 80%. As a result, such representative door-to-door travel time estimates can be critical to designing, planning, and engineering effective and sustainable urban mobility systems. Given the required data, the framework presented here is transferable to different routing engines, modes, data sources, and locations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient and representative door-to-door travel time estimation for planning and policy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月4日

查看全文
  • Marco Miotti,
  • Stefanie Hellweg

摘要

Travel times for many individual origin–destination pairs can be estimated without relying on detailed transport simulations by using commercial mapping services or by combining open-source routing engines with open street network data. While the former are more accurate, the latter are more scalable and economical. Neither option typically accounts for the time it takes to get to and unlock a vehicle or find parking. Here, we present a data-driven and scalable approach to obtain accurate, representative door-to-door travel time estimates for walking, cycling, e-bikes, and driving using open network data. To do so, we first use the open source routing machine (OSRM) to produce baseline estimates, then update these estimates based on characteristics of the built environment along the route and near the origin and destination of each trip. To apply this approach, we harness multiple travel survey datasets from Switzerland and evaluate the predictive performance of two statistical models: a linear regression and the XGBoost machine learning algorithm. We find that both models are capable of producing substantially more realistic travel time estimates than the baseline routing engine alone, achieving correlations with the Google Maps API of over 98%. Key predictors include intersection count and density along the route. Topography, critical for active modes, can effectively be incorporated using a constant time penalty per meter of elevation gained and lost. XGBoost is 5–20% more accurate than the linear model, depending on the mode. On the other hand, the coefficients from the linear model can be used to update the original routing engine to produce representative door-to-door travel time estimates without subsequent statistical processing. Finally, we highlight the importance of such estimates: Without adjustments, pedestrians and even bicycles are estimated to be faster than cars less than 3% of the time. With our adjustments, during peak hours and in or nearby cities, bicycles are faster than cars for 60% of all trip legs, and e-bikes for up to 80%. As a result, such representative door-to-door travel time estimates can be critical to designing, planning, and engineering effective and sustainable urban mobility systems. Given the required data, the framework presented here is transferable to different routing engines, modes, data sources, and locations.