R. Salih-Elamin, H. Al-Deek
In the last decade Bike share systems (BSS) have seen tremendous growth across the globe. The objective of this paper is to predict short-time and short-distance BSS trip duration throughout the day under weather conditions. Short-term prediction of BSS trip duration is important especially when the trip includes park and ride, and/or when it is coordinated with public transit. Updating BSS travel times frequently within the day (e.g., on a half-hourly basis) and in advance will help save time and money for both bike share users and system operators. In this paper, historical BSS trip travel time of a hundred capital bike share stations in Washington D.C. were modeled using several different modeling techniques: Stepwise Multiple Linear Regression (MLR), Autoregressive Integrated Moving Average (ARIMA), and ARIMA with exogenous variables (ARIMAX). The data was grouped into two datasets based on trip distances: the first group is for trips that take less than 0.5 mile, and the second group is for trips between 0.5 mile and 1 mile. The results show that temperature, fog, and distance between bike stations have significant effects on BSS travel time. Based on statistics of fit, Stepwise MLR model had a better performance and was chosen to predict travel times for the bike share system. A unique contribution of this paper is to provide a finer resolution prediction of BSS duration throughout the day under the effect of weather conditions. The results of this research are beneficial to bikers in pre-planning their trips, and to bike share system managers and operators in predicting travel times, determining bikes’ availability, re-allocating bikes, and relocating bike stations in the bike share network.
Keywords: bike share; stepwise multiple linear regression; travel time prediction; ARIMA; ARIMAX