Predicting Demand of Distributed Product Service Systems by Binomial Parameter Mapping: A Case Study of Bike Sharing Station Expansion

Abstract

Quantitative approaches for estimating user demand provide a powerful tool for engineering designers. We hypothesized that estimating binomial distribution parameters n (user population size) and p (user population product affinity) from historical user data can predict demand in new situations for distributed product service systems. Distributed product service systems allow individuals to use shared products at different geographic locations as opposed to owning them. This approach is demonstrated on a major bike-sharing system (BSS) expansion. BSSs position rental bikes around a city in docks at prescribed locations. BSS operators must predict the rider demand when sizing new docking stations, but current demand estimation methods may not be suitable for distributed systems. The main contribution of this paper is the development and application of a revealed preference demand estimation method for distributed product service systems. While much current research seeks to solve distributed system operational problems, we estimate the user population characteristic to provide insight into the initial installation design problem. We introduce the use of spatial surface plots to extrapolate binomial parameters n and p over the service area. These surfaces allow more accurate prediction of relative ridership levels at new station locations. By utilizing Spearman's rho as a comparison benchmark, our approach yields a stronger correlation between our prediction and the observed new station utilization (rho = 0.83, stations = 46, p < 0.01) than the order implemented by the BSS operator (rho = 0.59, stations = 46, p < 0.01).