A multiagent simulation framework for evaluating bike redistribution systems in bike sharing schemes

Abstract

In the recent years, more and more cities implement bike sharing schemes, where a set of bike is available for short-term rental between an origin and a destination station. Those systems are seen as an effective way to improve the image of biking in a city, but come with operational challenges. One particularly difficult problem is bike redistribution. Indeed, asymmetry in the travel demand and aversion for uphill riding often leads to shortage of bikes in stations with lots of departures and shortage of available slots in stations with lots of arrivals. This leads to the necessity to design redistribution strategies, usually by having light trucks take bikes from some stations and put bikes in others. The design of redistribution strategies to allow the provision of a reasonable Level of Service has mainly been studied under the angle of optimization, using historical data to create a model of demand. However, none of those approaches contains a model of the reaction of the demand to a change in the quality of the redistribution process. This paper thus takes the problem from a different angle: rather than designing optimal relocation strategies given a demand, it proposes a method to estimate the reactions of demand to a change in the relocation strategy. It demonstrates the usage of multi-agent activity-based simulation to estimate the reactions of demand to changes in the relocation strategies, taking into account explicitly the fact that a more efficient redistribution strategy will lead to a higher level of service, generating a new demand that might change the quality of the redistribution. This simulation system is tested on a scenario for the Zurich area, Switzerland, analysing the reaction of demand to two ideal redistribution strategies, i.e. no redistribution and optimal redistribution. Simulations show a big difference in demand depending on the quality of the redistribution, and even though no specific behavioral parameters for bike sharing were included, they seem to be able to reproduce the most salient characteristics of bike sharing travel from structural constraints only.