Improving the Service Quality of Bike Sharing Systems via the Analysis of Real-Time User Data

Over the last few years, bike-sharing systems (BSS) have become a cost-effective, environmentally friendly alternative for public transportation in many cities around the world 

Bike sharing is an innovative urban transportation alternative that provides citizens fast access to bicycles for inner-city commuting. Over the last few years, bike-sharing systems (BSS) have become a cost-effective, environmentally friendly alternative for public transportation in many cities around the world. From its early emergence about a decade ago to its consolidation in recent years, bike sharing (BS) has rapidly become a mobility paradigm proven to bring significant societal benefits for both its users and the cities adopting them. The provide citizens with a fast mode for making short commuting trips, a vibrant option for tourists to visit numerous city landmarks and attractions and an efficient way for riders to traverse cities in a healthy and exciting way. Similarly, from the perspective of the cities, BSS have become a natural platform for reducing traffic and pollution, for promoting healthy habits within its citizens, and overall, for improving the living conditions of the city.

For bike sharing systems to be considered effective transportation alternatives, they must provide their users with a high-quality service in the form of coverage and bike accessibility. While other transportation systems are readily available (e.g., personal vehicles) or follow predetermined schedules (e.g., bus or metro systems), BSS require to maintain a steady number of available bikes distributed across the service areas, so that potential riders have the possibility to use the system. Providing an acceptable service quality in the form of coverage and bike availability requires the operator to keep detailed track of customer travel behavior and demand patterns, which often poses a heavy operational load.

Planning the logistics of bike redistribution is a remarkably difficult task for it requires not only to identify optimal ways to effectively transfer batches of bikes between stations, but also because it depends on the operator's ability to generate consistent demand patterns and travel behavior estimates. Currently, most system operators use static models that are strictly based on historical demand averages. This type of models typically fail to consider valuable user information that is dynamically collected by the system via online queries.  Periodic and reactive bike redistribution trips are then exclusively performed when bike stations are deemed near empty or at full capacity, but are rarely planed and guided by predictive models with the ability to anticipate bike shortages given customer behavior.  With new technological developments rapidly permeating most bike-sharing systems (e.g., GPS tracked trips, online-locking and paying systems, and smart data collection via smarthphone apps), predictive analytic models have the potential to dramatically change the way in which bike sharing systems are operated.

The principal objective of this project is to develop a predictive statistical framework to efficiently estimate the ability of a bike-sharing system to serve incoming bike requests. By mining user data collected from the system's smartphone app, an operator can utilize the proposed models to predict the likelihood that any potential user who desires to use the system decides to do so under the given the system's conditions the user encounters (e.g., the location from where a bike request originates and its proximity to the nearest available bike, the weather conditions, the time of the day, and the customer pro_le, among others). The proposed statistical models is coupled with an operational bike redistribution model to analyze the cost-effectiveness of triggering quick bike redistribution tours to raise the service quality of the system to desired levels. The project also  investigates variations of the proposed models to support systems operated by either fixed stations or free-floating zones and will compare their efficacy given different demand scenarios.

This will enable the operator to:

  • better plan the redistribution process, focusing on areas of the system with larger probabilities of producing user rejections;
  • fine tune the OD demand estimations to better inform tactical decisions regarding the location of bike stations and the boundaries of free-floating zones;
  • monitor and assess the service quality of the system given various operational strategies, particularly during high-demand events.

It is important to note that methods are general in the sense that they are not limited to shared mobility systems. Similar type of models can be used of other systems, like electric-car rentals, and online transportation networks such as Uber and Lift.

Data: The proposed predictive statistical framework will be applied over data collected by the Reddy Bikeshare (RB) System, which currently serves the city of Buffalo, New York. RB operates a system with a fleet of 200 \smart" bicycles powered with GPS capabilities located at 35 stations throughout the city. During its operational months from mid Spring to late Fall, RB registered a total of 11,986 trips covering a staggering 17,614 miles.