Excess demand prediction for bike sharing systems

Abstract

One of the most crucial elements for the long-term success of shared transportation systems (bikes, cars etc.) is their ubiquitous availability. To achieve this, and avoid having stations with no available vehicle, service operators rely on rebalancing. While different operators have different approaches to this functionality, overall it requires a demand-supply analysis of the various stations. While trip data can be used for this task, the existing methods in the literature only capture the observed demand and supply rates. However, the excess demand rates (e.g., how many customers attempted to rent a bike from an empty station) are not recorded in these data, but they are important for the in-depth understanding of the systems’ demand patterns that ultimately can inform operations like rebalancing. In this work we propose a method to estimate the excess demand and supply rates from trip and station availability data. Key to our approach is identifying what we term as excess demand pulse (EDP) in availability data as a signal for the existence of excess demand. We then proceed to build a Skellam regression model that is able to predict the difference between the total demand and supply at a given station during a specific time period. Our experiments with real data further validate the accuracy of our proposed method.

Practical method to improve usage efficiency of bike‐sharing systems

“…Another study evaluated a log–log regression model for bike prediction. In addition, Liu and Pelechrinis [18] adopted a Skellam regression model to estimate excess demand, for example, how many users attempted to rent a bike from an empty station, and Gast and others [25] attempted to model bike availability in a station using a queuing model.…”

“…The second category involves construction of a prediction model. Various machine learning-based approaches have been proposed to construct effective bike rental and return prediction models [16][17][18][19][20][21][22][23][24]. For example, previous studies [19,21] have used long short-term memory (LSTM)-based models because LSTM can capture the long-term dependencies between data.…”

“…Many other approaches [18,22,24] have used traditional machine learning models to predict bike demand. For example, in a previous study [24], five prediction models, that is, CUBIST, regularized random forest, classification and regression tree, K-nearest neighbors, and conditional inference tree models, were evaluated to predict bike demands.…”

A Demand-Centric Repositioning Strategy for Bike-Sharing Systems

“…In addition, some studies focus on the prediction of rental and return demands for stations, while common models in recent years include multiple regression analysis [ 16 ], random forest [ 17 , 18 ], boosting framework [ 17 , 18 , 19 ], deep learning [ 18 , 20 ], etc. Several works employ probability distributions to model the number of trips at each station, containing negative binomial [ 17 , 21 ], Weibull [ 22 , 23 ] and Poisson [ 16 , 17 ], where the latter is often the best choice for this task. The demand forecasting can encourage operators to grasp the urgent bike/stall demand of stations and reallocate bikes accordingly.…”

“…Contardo et al [ 27 ] firstly modeled the dynamic repositioning problem as an optimization problem on the complete directed graph, and then proposed two decomposition schemes to obtain feasible solutions in tractable time. Based on the demand estimation by a stochastic process, e.g., Poisson [ 16 , 17 ] and Markov chain [ 28 , 29 ], some works can significantly reduce the complexity of the prediction problem, where Liu and Pelechrinis [ 16 ] adopt the Poisson regression and Skellam regression to estimate the excess demand of bikes or stalls. Chiariotti et al [ 6 ] give an alternative route of the Birth–Death process to model stations’ occupancy and estimate the amount of time a station is unavailable, i.e., either empty or full.…”