Bike sharing systems are very popular in large cities around the world. The user checks out a bicycle at a nearby station and returns it in a station nearby the destination. For smooth functioning of the system, both bicycles and slots should be available at every station. Otherwise, users have to look for other nearby stations and this will make them avoid using the system. The flow of bikes to and from a station will not follow a similar pattern. This will cause imbalances in the spatial distribution of bikes. There are many ways of managing this problem, such as shifting the bikes from one station to another by trucks and locating stations by analyzing the demand, but all of this depends on understanding the spatiotemporal bike demand. This article shows that bike demand at each station can be modeled as temporal point process considering bike checkout times as random events. The data on Divvy Systems of Chicago is publicly available and can be analyzed to understand the demand pattern. The data analyzed pertains to the period between April 1 and November 30, 2016. This date range was selected since bike usage reduces considerably during the winter. During this period, there were 3,068,211 bike trips and 458 active bike stations. Depending on the location, there is variation in demand. It varied from 29 annual trips in station 386 in the south side to 85,314 annual trips for station 35 at the Navy Pier. Demand for stations with large counts can be estimated by kernel smoothing or other density estimation methods. However, for stations with low daily counts, this method will not work well. This article presents a method to overcome this problem by borrowing data across different days. This can be used to study spatial correlations between stations and thereby clusters can be identified.
展开▼
