Stochastic Bike-Sharing Transport Network Design

As a transport mode, bike-sharing has gained popularity worldwide because it is environmentally friendly and cost-efficient. However, as a bike-sharing network grows, operating costs at rental centers increase. The problem is determining the locations of rental centers to open and the number of bicycles that will be transferred daily between rental centers while minimizing the total operating costs. We present a stochastic programming model and a Benders decomposition-based hybrid algorithm. We consider two scenarios for demand-return machine learning models - time series-based prediction and weather-based forecasting. Finally, we provide a case study of developing a bike-sharing network in New York City to verify the significance of the proposed models. We also evaluate the performances of demand-return prediction models and the impact of the relative ratio between demand and return on bike-sharing network design. We find no bicycle transfer if the penalty cost for a rental station has an inverse linear relationship with the ratio of returns to rentals. Nevertheless, when the penalty cost is exponentially dependent on the negative ratio of returns to rentals, bicycle transfer occurs between rental stations with large ratios.