Microtransit deployment portfolio management using simulation-based scenario data upscaling

Abstract

Due to transportation technologies having such heterogeneous impacts on different communities, there needs to be better tools to evaluate the deployment of emerging technologies with limited data. Microtransit is one such technology. We propose a novel framework based on existing methods to "upscale" the limited data available so that further decision-support analysis and forecast modeling can be achieved where none could prior. The framework involves expanding an initial day-to-day adjustment process to handle both first/last mile access trips and direct trips, updating a within-day microtransit simulator with a parametric design, and developing a synthetic scenario generation process. The framework is tested in a case study with data from Via for Salt Lake City, Austin, Cupertino, Sacramento, Columbus, and Jersey City showing an average 18% ridership error for the market equilibrium models. Data from four of those cities are upscaled to 326 synthetic scenarios to estimate forecast models for ridership and fleet vehicle-milestraveled using Lasso regularization. While the models have root mean squared error (RMSE) values between 37-45% of the averages, using only four cities' data alone would not produce any forecast model at all. The results show that variables with statistically significant positive impact on ridership and negative impact on vehicle-miles-traveled (VMT) include zones with more transit stations, higher employment, but lower "employment density x fixed fare". The models are then used to identify two alternative portfolios with similar fleet VMT as the original four cities but are forecast to have up to 1.9 times the ridership.