Modelling within-day ridehailing service provision with limited data.

This paper proposes a holistic, within-day ridehailing service provision modelling approach, focusing on large-scale services with human-driven fleets. The model is powered by a full, two-year record of ridehailing trips in Toronto. Canada. Despite the abundant data, operations and fleet-related data still lacks, and uncertainty still exists concerning actual service provision deployment. Hence, a novel approach dealing with limited data availability is developed, which extracts implicit supply levels from trip-based demand data. The model is time-step driven, agent-based, and uses list-based data structures. It focuses on three key components: matching mechanisms, time intervals, and (several features of) driver activity. After testing various component combinations, a Hungarian matching algorithm, a normal distribution of driver work hours, and 5-minute time intervals allow close replication of observed wait time distributions and unique drivers per hour. Emergent model outputs also include VKT by vehicle states (idling, en-route, in-service), vehicle shifts and trip chains. [ABSTRACT FROM AUTHOR]