Shared Mobility-on-Demand Systems

Abstract

On-demand ridesharing might be an effective solution to reduce traffic congestion and, as a result, reduce emissions to mitigate environmental issues. However, travel times are more uncertain in ridesharing services compared to non-shared transit services due to their shared nature. For example, at one moment, the users are informed that the travel time takes five minutes, a moment later, a request is added to the trip and the users are now confronted with a travel time of ten minutes instead. This uncertainty in travel time is called unreliability. Unreliability is perceived as inconvenience by the users and it has a negative impact on the users' travel habits regarding ridesharing service.

Therefore, this research provides a novel solution to deal with this problem while mitigating the negative effects on the other quality-of-service indices, e.g., waiting time, rejection rate, and delays. This is achieved by providing two indicators to each new user: (1) the possibility that an additional request will be added to the trip that negatively influences the trip duration of the user; and (2) a prediction of the magnitude of a possible delay for whenever the case occurs that a new request will be added to the trip. These indicators are based on predictions, created by shareability shadows which determine the number of requests that might be added to the trip, and implementing a simple time-series demand forecasting method named exponential smoothing. The shareability shadow makes predictions by confining the regions of ridesharing opportunities by analyzing the travel constraints of the involved users. The solution is validated using a state-of-the-art routing and assignment method and a test case of Manhattan, New York City. With the formulated measure, on average, 58% of the requests receive a correct prediction about the possibility of a negative change, 17% receive an indecisive prediction, i.e., a prediction stating that the possibility of a negative change is "medium", and 25% of the requests receive an incorrect prediction about the possibility of a negative change. Moreover, the measure successfully provides a predicted magnitude of the travel time increase with a 95% confidence interval.