Fig 1.
(a), (b) In-passenger-flow Nin(s,t) and out-passenger-flow Nout(s,t) of each subway station s during the time window 9:00 a.m.–9:15 a.m. of a typical weekday in 2014.
Fig 2.
Distributions of passenger-flow show heterogenous patterns in different subway stations. (a) is for in-passenger-flow Nin(s,t), (b) is for out-passenger-flow Nout(s,t).
Fig 3.
Anomalous passenger flow detection for “Window of World” subway station.
(a) Out-passenger-flows of the “Window of World” subway station over the whole observation period. Anomalous flows and ordinary flows are discriminated by the green line fε. (b) Red circles represent anomalous out-passenger-flow of the station during the time window 7:00 p.m.–7:15 p.m., while blue circles represent passenger flows under ordinary traffic conditions.
Fig 4.
Predictive results versus ground-truth data for subway station “Window of World” when nstep = 1.
(a) Results for a typical weekday. (b) Results for a day when mass events occurred near the station.
Table 1.
The average error of the four models when nstep = 1.
Fig 5.
Performance of four models when predicting out-passenger-flow records in subway station “Window of World” on December 30, 2014.
(a-h) When the prediction was made 1 to 8 time windows ahead of the target time window, respectively.
Fig 6.
Performance of four models when predicting out-passenger-flow records in subway station “Window of World” on December 31, 2014.
(a-h) When the prediction was made 1 to 8 time windows ahead of the target time window, respectively.
Table 2.
The average error of the four models.
Fig 7.
Performance analysis of the predictive models.
(a), (b), (c) Performance of the four predictive models under the ordinary out-passenger flow condition for all stations during the whole test period. (d), (e), (f) Same as (a), (b), and (c), but for the results under the anomalous out-passenger flow condition.
Table 3.
The average error of all in-passenger-flow of all stations.
Table 4.
The average error of all Out -passenger-flow of all stations.
Table 5.
The average error of all in-passenger-flow (nstep = 2).
Table 6.
The average error of all Out -passenger-flow (nstep = 2).