Can Data Mining Help Car Sharing?
Chiara Boldrini, CNR, Raffaele Bruno, CNR, Haitam Laarabi, CNR
We provide a spatiotemporal characterisation of how customers use a car sharing service, relying on the analysis of two datasets comprising the pickup and drop-off events in two large car sharing systems (one station-based, one free-floating).
Mobility and congestion are critical concerns for every city, be it large or small, due to the economic and environmental challenges that they pose. Many analysts have advocated addressing these issues by encouraging new multimodal services and fostering the deployment of more efficient on-demand mobility services. In this work, we focus on car sharing, a mode of transportation that is gaining increasing popularity with its promise to reduce traffic congestion, parking demands and pollution in our cities. There are two main classes of car sharing services: station-based car sharing or free floating car-sharing. In the former, shared vehicles are picked up and dropped off at designated (and reserved) locations within the service area, called stations. In free floating car sharing, instead, cars can be picked up and dropped off anywhere within the service area, as long as parking is permitted at that location. The two approaches have each advantages and disadvantages. Free floating car sharing offers a lot of flexibility to customers, but in cities were finding a parking spot is troublesome, the reserved parking space offered by station-based car sharing may appeal more.
Managing a car sharing service is generally costly. For example, there are the costs for the infrastructure (high especially for station-based car sharing) and there are the costs for optimising the car sharing operations. There is an ongoing discussion within the car sharing community about the opportunity and convenience of vehicles redistribution. The problem that motivates redistribution is that the car sharing system tends to become unbalanced during the day, with cars that get stuck in so-called cold spot while they would be needed in hot spots. The solution would be to move these unused cars from cold spots to hot spots. Unfortunately, the cost of redistribution can be significant (two operators are needed for each redistributed vehicle) and it is crucial to perform it in an optimal way. For redistribution, and for the car sharing operations in general, we advocate the use of real car sharing data, in order to uncover, quantify, and model properties of these systems that could be used to perform this optimisation of the car sharing operations.
The goal of this work is to provide a spatiotemporal characterisation of the car sharing system, and to compare and contrast the usage patterns of two real-life car sharing systems offering a different type of car sharing service (station-based vs free floating) in two different cities. First, we discuss how the two systems can be compared on equal grounds, despite their intrinsic differences. Then, we provide an aggregate view of the temporal evolution of system usage during the day. Finally, exploiting a clustering technique, we show how the heterogeneous usage patterns (i.e., how customers pick up and drop off vehicles) at individual stations can be expressed in terms of only two classes of behaviours. This clear dichotomy is linked to the customers’ typical daily life: thus, we find that there are stations that attract cars mostly in the morning and stations attracting cars mostly in the evening, depending on the nature – residential or business – of the area. Being able to identify the class to which each station belong can be crucial for vehicle redistribution: cars should never be moved between stations of the same class. In fact, these stations tend to experience peak demand at roughly the same time of the day, hence it would be like moving a car from a hot spot to another hot spot, which is not desirable.
Association for European Transport