OVERVIEW
Challenge
How to intuitively assist rebalancing issues related to “Last Mile Problem” for bike sharing programs in China?
Timeline
8 weeks
Project Info
Individual project
Solution
I design MobikeViz to visually analyze temporal and spatiotemporal patterns of popular free floating bike sharing system (FFBSS) Mobike in Shanghai. Mining of over 32 million data points, I found strong cyclical variations on temporal patterns of usage and analyzed spatiotemporal patterns through data visualization encoded by a geohash index based spatial data mode.
Mentors
Ercument Gorgul
Patent Number
201910653052. 1
Outcome
Research
Background survey
Historically, there have been four generations of public bike sharing systems. Starting with free bikes in the 60’s and coin operated bikes in early and mid-90’s. Past decades’ advancements in technology led to third generation of systems - docked bikes - that can be checked in and out from their designated stations electronically. This enabled first time, collection, storage and analysis of real-time usage data possible. Built upon the technology of third generation systems are the demand-responsive fourth generation systems that are dockless, traceable via GPS and fully integrated with real-time technology and infrastructure .
First generation
“Free bikes“
Second generation
“Coin-deposit systems”
Third generation:
“IT-based systems”
Fourth generation:
“Demand-responsive/
multimodal systems”
Currently Asia represent the world’s fastest growing bikeshare market among Europe and America. With over 61,000 bikes and 2500 stations, China constitutes the majority of that growth. Much of this expansion is backed by Chinese government’s policies in order to reduce emissions. A large number of cyclists, extensive and sophisticated bicycle transportation infrastructure in almost all big Chinese cities with separated bike lanes bicycle, on and off-road bicycle parking, as well as history of bike use habit might have contributed to this growth.
A travel survey conducted in Beijing illustrate that 71.6% of all bicycle trips in the city cover less than five kilometers, signaling the demand of “last mile problem”, indicating the starting/ending node of a trip before and after a public transit stop.
To serve such demand as much as possible, it is significant for public shared bike operators to detect the usage patterns from temporal and spatial distribution, e.g., when will people choose to use shared bikes and are there any routine or anomaly caused by any reason for user behavior. This type of information enable operators effectively rebalance the distribution of shared bike among locations.
What is rebalancing shared bike？
As a solution for the first/last mile problem in many cities with public mass transit systems, FFBSS can provide means of transportation between existing hubs of these systems and desired destinations.
Due to one-way and short-time usage, highly dynamic spatiotemporal movements cause imbalances in the distribution of bikes. Thus, the rebalancing of shared bikes is a major problem since it could be costly for the operator both from organizing the distribution of bikes as well as well as catering to demand.
Visualization
This section will cover the methodical steps of the visual analysis from temporal and spatial perspective.
Time-based visualization
As first step of visualization, a heat map is generated based on Mobike request data to find what days of the week and what time of the day have more frequent use of bikes. Heat map, revealed a result that can be assumed: High traffic during the morning and evening rush hours (from 7 to 8 am and from 5 to 7 pm). It is also worth to mention that variance of bicycle demand during work day is greater than that of during weekend.
To further compare the change between days and visualize details on similar patterns as well as anomalies a line chart is generated by using start time as the row and amount of requests in each minute in one week as the column
Our main finding includes:
· Curves on weekend are similar to each other and have some unique patterns: All of them have three peaks: 7 - 8 a.m., 11-12 a.m., 5 - 7 p.m. Peaks during 5 - 7 p.m. and are higher than peak of 11-12 a.m.
· Curves on weekday have similarities to weekend curves with three peaks around same hours of weekend except for the day of March 22nd. Peak during 7 - 8 a.m. is higher than peak of 11-12 a.m. and 5 - 7 p.m. peak of is higher than the one during 7 - 8 a.m.
· Curves on weekday are higher than those on weekend, representing higher volume of use. For example, curves between 7 to 8 a.m. during weekdays are almost twice than those on weekend.
Location-based visualization
After analyzing temporal patterns, we turn to look into spatial factors. To validate the first/last mile problem as one of the main uses of FFBSS, it is necessary to compare trip distances. I used geohash to process our data.
Geohash is a public domain geocoding system, which encodes a geographic location into a short string of letters and digits. It is a hierarchical spatial data structure which subdivides space into buckets of grid shape. Geohashes offer properties like arbitrary precision. The longer a shared prefix is, the closer the two places are. In other words, this index structure can be used for a quick-and-dirty proximity search: the closest points are often among the closest geohashes.
I generated a 7-character string that enabled easier assessment of the distance by comparing the similarity of these two geohash strings.
The figure reveals the emergence of a linear pattern, confirming the main use of Mobike within short distances. More specifically, this is due to alphabetical proximity of geohash values for pickup and dropoff locations. If 5 characters of the 7-character geohash is same, the ride will be plotted close proximity to blue line. Data in this diagram is informing 95.47% bike rides are less than 3 km distance.
One of the most important discovery of this visualization is the emerging rectangular vacant patterns along the line. The left and bottom sides of these rectangular formations made out of intensified plotting of the points of pickup and dropoff respectively, indicating the check-in and -out activity in those areas. Larger the size of these rectangular formations, more frequent Mobike trips from this area.
Discussion
The purpose of this study was to develop a visualization that can help to provide more accurate understanding the locations of high demand and provide and efficient tool for rebalancing. The basis of our contribution stands on our effort to find a novel way of visualize bike sharing data by preprocessing the spatiotemporal information, so it can be used not only to understand the trend but also to inform operations.
Specifically, the findings of our study are as follows:
Firstly, the fact that the use of Mobike is more frequent during weekdays means the major demand of Mobike is to facilitate commuting between workspace and residence.
Second, during weekend, the highest peak is in the evening while during the weekday the highest peak is in both the morning and afternoon, which means people tend to stay at home during weekend morning and use shared bike during the evening.
Finally, during weekdays, use of Mobike at noon is more likely to be affected by sudden change of external factors and events than that in the morning or evening.
This become evident for the data on May 22nd with the disappearance of and expected peak at noon time. A quick research on historic weather and temperature record of the area revealed a rainfall and sudden cooling of temperature around noon time, providing weather conditions as a plausible reason for anomaly. Another one was for the sudden shift between weekend and workday flows (i.e. more traffic in weekend and less in weekday). The reason for this was the arrangement made by the Chinese government for the occasion of the highly celebrated dragon boat festival holiday, as a day of the weekend prior to holiday is swapped as a working day in order to consolidate a normally one-day holiday, into a 3-days holiday (Saturday as a workday and holiday being from Sunday through Tuesday).