Navigating Widespread Urban Transit Dynamics with Standardized Data and Scalable Models

Abstract

Standardized and open source bus data including static schedules and realtime positions have become widely available in public web application programming interfaces. Though these data primarily underly popular mobile trip planning applications, they also enable new analyses in understanding, forecasting and improving bus operations across cities. Due to their lower resolution and simpler features, open data are more challenging to work with than those of the underlying sensors. However, their wide scale and standardization make them a valuable resource for researchers and planners. This work develops a set of tools for analyzing bus operations with open data. Central to this endeavor is the ongoing collection of a multi-year dataset from the King County Metro transit network in Seattle, Washington, approaching one billion tracked bus locations. First, basic roadway segment aggregation is used to visualize the spatiotemporal dynamics of different delays in the transit system. This is used to identify priority locations for transit priority treatments. Then a set of deep learning models are developed to forecast bus travel times under different data availability scenarios. Their generalizability is tested across different cities and transit networks. Finally, these models are used to estimate energy demands of a battery electric bus fleet for any city. Implications of the open data standards for energy modeling are examined, and a cross-sectional analysis reveals barriers to fleet electrification.