data used in this study were collected from the 
Beijing subway between 05:00 and 23:00 for five consecutive 
weeks from February 29 to April 3, 2016. There were 17 lines 
and 276 subway stations (excluding the airport express line and 
the stations on it) in March 2016 in Beijing. Only data from the 
25 workdays in the target period are applied to this study, 
constituting 130 million records. Every record contains the card 
number, entry-station number, exit-station number, entry time, 
exit time, entry-station name, and exit-station name. Inflow and 
outflow time series are extracted according to (7). The time-granularity TGs
used in our study are 10, 15, and 30 min. Therefore, it is easy to 
integrate the results predicted for TG = 10 and 15 min into those 
predicted for TG = 30 min to compare prediction performances
for different TGs