[Paper Review] Balancing bike sharing systems (BBSS): instance generation from the CitiBike NYC data

Tommaso Urli|arXiv (Cornell University)|Dec 13, 2013

Urban Transport and Accessibility2 references5 citations

TL;DR

This paper proposes a method to generate realistic, static benchmark instances for balancing bike sharing systems (BBSS) using real CitiBike NYC data. By analyzing midnight snapshots from September 2013, the authors derive station bike counts, target levels, and inter-station distances via Google Maps, then generate 180 diverse instances of increasing size (40 to 240 stations) with balanced source and sink stations, enabling reproducible optimization testing.

ABSTRACT

Bike sharing systems are a very popular means to provide bikes to citizens in a simple and cheap way. The idea is to install bike stations at various points in the city, from which a registered user can easily loan a bike by removing it from a specialized rack. After the ride, the user may return the bike at any station (if there is a free rack). Services of this kind are mainly public or semi-public, often aimed at increasing the attractiveness of non-motorized means of transportation, and are usually free, or almost free, of charge for the users. Depending on their location, bike stations have specific patterns regarding when they are empty or full. For instance, in cities where most jobs are located near the city centre, the commuters cause certain peaks in the morning: the central bike stations are filled, while the stations in the outskirts are emptied. Furthermore, stations located on top of a hill are more likely to be empty, since users are less keen on cycling uphill to return the bike, and often leave their bike at a more reachable station. These issues result in substantial user dissatisfaction which may eventually cause the users to abandon the service. This is why nowadays most bike sharing system providers take measures to rebalance them. Over the last few years, balancing bike sharing systems (BBSS) has become increasingly studied in optimization. As such, generating meaningful instance to serve as a benchmark for the proposed approaches is an important task. In this technical report we describe the procedure we used to generate BBSS problem instances from data of the CitiBike NYC bike sharing system.

Motivation & Objective

• To create realistic, reproducible benchmark instances for balancing bike sharing systems (BBSS) to support optimization research.
• To address the lack of standardized, publicly available BBSS instances derived from real-world data.
• To ensure generated instances reflect real-world dynamics such as peak usage patterns and spatial imbalances.
• To enable fair comparison of optimization algorithms by providing consistent, data-driven problem instances.
• To support the development of efficient rebalancing strategies through a scalable, open-source instance generator.

Proposed method

• Collected 25,000 CitiBike NYC snapshots at 10-minute intervals from May to November 2013, focusing on working days.
• Computed hourly distributions of bike counts per station to identify sources (outflow) and sinks (inflow) using first and third quartiles.
• Defined a displacement value $ disp_s = ig floor C_s - (C_s - (max_s - min_s))/2 ig floor - max_s $ to minimize the risk of empty or full stations.
• Selected midnight snapshots from September 2013 as the starting point for instance generation, assuming rebalancing occurs overnight.
• Chose station ID 294 as the depot due to its central location and excluded it from station selection for other instances.
• Generated instances by randomly selecting stations from source and sink groups to balance the objective function range and ensure diversity.

Experimental results

Research questions

• RQ1 How can realistic, static BBSS problem instances be generated from real-world bike sharing data?
• RQ2 What are the temporal and spatial patterns of bike imbalance in a major urban bike sharing system like CitiBike NYC?
• RQ3 How can station-level bike counts and target levels be derived from historical usage data to reflect real-world imbalances?
• RQ4 What is the impact of depot selection and station sampling strategy on the diversity and realism of generated instances?
• RQ5 How can inter-station distances be accurately modeled to reflect real travel times and routes in urban environments?

Key findings

• The authors successfully generated 180 static BBSS instances of increasing size from 40 to 240 stations using real CitiBike NYC data.
• Midnight snapshots from September 2013 were used as the baseline for initial bike counts ($b_s$) and target levels ($\hat{b}_s$), ensuring temporal realism.
• The generator uses a balanced sampling strategy that selects stations from both source and sink groups to ensure a broader objective function range.
• Inter-station distances were computed via the Google Maps API in both meters and minutes, providing accurate travel cost inputs.
• The generator is open-source under the MIT license and publicly available at https://bitbucket.org/tunnuz/citibike-nyc-generator.
• Unlike standard BBSS instances, this dataset does not include loading/unloading times in travel times, allowing flexible modeling of different operational policies.