Travel Behavior Analysis for Free-Floating Bike Sharing Systems Based on Markov-Chain Models

Abstract

The emergence of the Free-Floating Bike Sharing System (FFBSSs) has brought convenience to the public and also posed new challenges to urban construction and management. Inspired by the ability of Markov chains to handle large volumes of data in Google’s PageRank algorithm, we propose a Markov-chain based approach to model the FFBSSs for capturing its macroscopic aggregated properties. The geohash based algorithm is proposed to divide a geography map into cells due to the non-stock feature of the FFBSSs. After this, the transition matrix of the Markov chain is built based on historical bike trip data. Spectral clustering properties and the characteristic that Kemeny constants can identify the critical regions are discussed. Then we use about 3.2 million bike trips real data of BJUT Beijing, China from Mobike to demonstrate its application in identifying clusters and critical stations. In our empirical study, three clusters are identified in the vicinity of the BJUT, one of which is further analyzed and then 10 critical cells corresponding to the major sites in the cluster are identified, which is in line with reality.