User Behavior Analysis And Optimization In Bike-sharing Systems

As an environmental-friendly and convenient mobility strategy,public bike-sharing has grown dramatically worldwide in recent years.It meets the public demand for short journeys and provides convenient,high flexibility and low-cost transportation.However,the advan-tages can not cover up the increasingly prominent issues.For stations,the user demand is ever-changing and unbalanced,which often leads to the check in or check out service un-available at some stations and has negative impact on user experience.For bikes,the usage frequency of each bike is unevenly distributed,and a small part of bikes are used much more frequently than others.Bikes that are used too much are vulnerable and hence increase repair bills and lead to potential service denied,posing a problem for both riders and system operators.At present,operators perform bike redistribution based on monitor video and user complaints,distribute bikes based on dispatchers' experience and set up a large number of staff to maintain and repair bicycles.However,these solutions only try to remedy problems after they arise.This paper,starting from massive data,extracts user behavior rules and bike usage patterns through data analysis and machine learning techniques.On the one hand,for user demand unbalance problem,we predict the user demand of each station to obatain the stock status in the future,so that operators can rebalance stations in advance.On the other hand,for bike usage unbalanced problem,we design a trip advisor framework to guide and optimize user check in and check out behaviors and balance bike usage frequency actively.It can then prolong the service life of bikes and reduce the system operating costs.Specific contributions are as follows:Firstly,we analyze the overall characteristics,distribution in time and space of user demand and bike usage distribution in the system after data preprocessing.The K-means algorithm is used to cluster the stations with similar demand patterns.Based on the anal-ysis,we can draw a conclusion that the primary problems in bike-sharing are unbalanced distribution of user demand and bike usage frequency.To alleviate the unbalanced demand problem,we establish a fine-grained demand fore-casting model and predict check in and check out demand on a per-station basis with sub-hour granularity by using random forest algorithm,which allows operators to make scientific and reasonable decisions in time.In our model,offline features such as time and weather are selected to capture the periodic patterns of user demand.Online feature is to reflect the real-time availability of the station which is helpful for abnormal traffic.To balance the usage frequency of bikes,we analyze the underlying reasons for the unbalanced usage and propose the concept of activeness.A novel data-driven utilization-aware trip advisor is designed to guide and optimize user behaviors,leading users to shift bikes between highly active stations and inactive ones,so as to balance the bike usage number and usage time.