You are on page 1of 6

Advances in Engineering Research, volume 193

2nd International Symposium on Transportation Studies in Developing Countries (ISTSDC 2019)

Calibration and Validation of Walking Behavior


Parameter (Case Study: Sky Bridge of Sultan
Mahmud Badaruddin II Airport, Palembang)
Siti Raudhatul Fadhilah Sony Sulaksono Wibowo
Magister Program of Civil Engineering Study Program of Civil Engineering
Bandung Institute of Technology Bandung Institute of Technology
Bandung, Indonesia Bandung, Indonesia
sitiraudhatulfadilah@gmail.com sonyssw@gmail.com

Abstract—A micro-simulation model is a crucial tool in the TABLE I. FACTORS I NFLUENCE THE SOCIAL FORCE MODEL
study of transportation, especially in complex traffic systems Factor Effect Description
that include interactions between components in it. Therefore,
Pedestrians take the shortest
the accuracy of the model must be considered by calibrating Desired direction Direction
path to their destination.
the parameters used and validating the model with observation Free-flow speed. Varies
data. In this study, the study area is the sky bridge which Desired speed Acceleration
between pedestrians.
connects LRT station and Sultan Mahmud Badaruddin II Within which time pedestrians
Relaxation time Acceleration
International Airport terminal, Palembang. The micro- try to reach their desired speed.
simulation model used is a model developed by PTV Group, Repulsive
Pedestrians avoid getting close
Viswalk. Calibration of pedestrian walking behavior is done by Other pedestrians to strangers. They have their
effect
trial-error to find the parameter value which represents the private spheres.
actual pedestrian behavior. While the model validation uses a Pedestrians avoid buildings,
Repulsive
Obstacles fences, etc. The closer they get
nonparametric statistical test on observational data. effect
the more attention they pay.
Attractive Pedestrians are attracted by
Keywords: Viswalk, walking behavior, calibration, Attractions
effect friends, window displays, etc.
validation Things in front of pedestrians
Weighting of
Angle of sight affect their motion a lot more
factors
than things behind them.
I. INTRODUCTION
Some randomness is included,
Random
Pedestrian density has become a common phenomenon Fluctuation
variations
e.g. for the choice between two
that must be considered, especially in public facilities, such equal options.
as train stations, airports, tourism areas, and sports stadiums. There are ten walking behavior parameters on Viswalk
This encourages various studies on pedestrians, both related that will determine how pedestrians move relative to other
to circulation, safety, and quality of service. To facilitate pedestrians and surrounding objects [4], as follows.
engineers in analyzing any types of pedestrian problems in
the field of transportation, many transportation experts 1) Tau (τ): This parameter represents the relaxation time
recommend using the micro-simulation model. or inertia that can be related to response time, as it couples
the difference between the desired speed and direction with
The microscopic simulation model allows analyzing a
complex system and being able to capture the interactive the current speed and direction for acceleration [4].
effects of different components of the system [1]. Moreover, 2) Lambda_mean (λm): This parameter governs the
the micro-simulation model is recommended for complex amount of anisotropy of the forces from the fact that events
traffic operations, significant conflicts between different road and phenomena in the back of pedestrians do not influence
users, road works, operation of dynamic control systems, him (psychologically and socially) as much as if they were
large-scale traffic studies, and public transport operations [2]. in his sight [4].
Considering the importance of the micro-simulation model in
transportation analysis, calibration of walking behavior 3) A and B social isotropic: Together with lambda,
parameters will be carried out by using the Viswalk program these parameters influence one of the two forces which form
in this study. Furthermore, the simulation results will be the repulsive force between two pedestrians [5].
validated referring to observational data.
4) A and B social mean: The parameter A social mean
decides the strength of the force, while the parameter B
II. LITERATURE REVIEW
social mean determines the range of the force in meters [4].
Several factors must be considered in the social force
model [3], which is the basis of the Viswalk program, as 5) Noise: This parameter determines the strength of the
shown in TABLE I. random force term. The greater this parameter value, the
stronger the random force that is added to the systematically
calculated forces if a pedestrian remains below his desired
speed for a certain time [4].

Copyright © 2020 The Authors. Published by Atlantis Press SARL.


This is an open access article distributed under the CC BY-NC 4.0 license -http://creativecommons.org/licenses/by-nc/4.0/. 46
Advances in Engineering Research, volume 193

6) React to n: This parameter sets the maximum number III. METHODOLOGY


of how many pedestrians are considered for the social force. The calibration process of pedestrian walking behavior
During the calculation of the total force for a pedestrian, parameters and model validation is shown by the following
only the influence exerted by the n closest pedestrians are flow chart in Fig. 1. To obtain pedestrian speed data on the
taken into account [4]. sky bridge, video recording is carried out by placing the
7) VD: Together with A and B social mean, this camera in the corridor and commercial area. The main
survey was conducted on Sunday, 2018, November 18, 15:30
parameter will determine the time required by pedestrians to
to 17:00, which is the peak hour of LRT passengers. The
avoid other pedestrians walking towards him [5]. pedestrian movements are recorded simultaneously at both
8) Side preference: This parameter defines whether points with a duration of 30 minutes each.
opposing pedestrian flows prefer using the right on the left Model validation will be done by using the chi-square
side when passing each other: -1 is for preference of the test to prove whether the observed frequency has a
right side, 1 is for preference of the left side, default 0 is no significant difference with the expected frequency. The
preference, behavior as before: pedestrians do not shun each indicator used is pedestrian walking speed in the sky bridge,
other [4]. precisely on the first corridor after the escalator access from
the airport terminal, as well as the commercial area, from the
9) Queue order: This parameter defines the orderliness
results of simulations and observations. There are several
of the queue formed by pedestrians, with values between 0.0
assumptions used in this study:
and 1.0. The higher this value, the more orderly pedestrians
line up one behind the order [4].  Pedestrians tend to choose the fastest route than the
shortest one.
Queue straightness: This parameter defines the shape of
the queue, with values between 0.0 and 1.0. The greater the  Pedestrians walking behavior in the first 30 minutes of
value, the more straight the queue will look [4]. observation is considered the same as the next 30
minutes.
 Pedestrians do not walk in groups but walk individually

Fig. 1. Methodology.

47
Advances in Engineering Research, volume 193

IV. DATA ANALYSIS AND DISCUSSION

A. Calibration of Walking Behavior Parameter


Calibration of pedestrian walking behavior parameters is
done by trial-error to find the appropriate value which
represents the actual pedestrian behavior. Therefore, it is
necessary to know the tendency of indicator towards
changes in walking behavior parameters compared to the
default settings of Viswalk based on research by [6]. The Fig. 6. The sensitivity of walking speed to A social mean.
figures below show the sensitivity of each walking behavior
parameters (see Fig. 2 to Fig. 10)

Fig. 7. The sensitivity of walking speed to B social mean.


Fig. 2. The sensitivity of walking speed to tau.

Fig. 3. The sensitivity of walking speed to lambda.


Fig. 8. The sensitivity of walking speed to noise.

Fig. 4. The sensitivity of walking speed to A social isotropic. Fig. 9. The sensitivity of walking speed to react to n.

Fig. 10. The sensitivity of walking speed to VD.


Fig. 5. The sensitivity of walking speed to B social isotropic.

48
Advances in Engineering Research, volume 193

Based on the evaluation of each walking behavior within a given distribution range is 13. The simulation results
parameter, the value that will be used in the next analysis can be seen in TABLE III.
shown in TABLE II below. If the simulation results with the
The average pedestrian speed in the table will be used for
application of these parameters do not yet represent the
observational conditions statistically, then it is necessary to model validation. The maximum average pedestrian speed
and minimum average density in the corridor and
change the value of walking behavior parameters. It is
assumed that all pedestrian types have the same behavior. commercial area are generated in the 13th simulation.

C. Pedestrian Volume
TABLE II. WALKING BEHAVIOR PARAMETERS
According to [7], “there is no existing guideline as to the
Parameter Default Value Value After Calibration modeling criteria that can be used for pedestrian calibration,
Tau 0.4 0.2 a recommended preliminary guideline adapted from [8] of 10
React to n 8 8 percent deviation from the observed platform pedestrian flow
Lambda 0.176 0.3 volumes and a 5 percent sum of all pedestrian flows in the
A Social (Isotropic) 2.72 1.3 network seems a reasonable accuracy”. TABLE IV to VII
B Social (Isotropic) 0.2 0.2
show the calculation of pedestrian volume deviations caused
by observation and microsimulation model carried out in the
A Social (Mean) 0.4 0.4
corridor and commercial area of the sky bridge.
B Social (Mean) 2.8 2.8
VD 3 6 From TABLE IV to TABLE VII, it is known that the
Noise 1.2 1.2 deviation of pedestrian volume between observations and
Side Preference None None microsimulation models is less than 10%. Thus, the
Queue straight 0.4 1
requirements stated in [8] have been fulfilled.
Queue order 8 1
TABLE IV. PEDESTRIAN VOLUME IN THE CORRIDOR (1)
B. Simulation Result
To the LRT Station
Determining the number of simulation runs is an essential
Simulation Run- Volume Deviation
step in developing a simulation model that is expected to
have data characteristics that represent the observational Observed Simulated Absolute %
conditions statistically. Errors generated by the simulation 1 67 62 5 8.06%
model are directly proportional to the number of repetitions 2 67 66 1 1.52%
performed. The number of simulation repetitions needed (n) 3 67 66 1 1.52%
is calculated by the following equation, where S is the 4 67 67 1 0.75%
average deviation standard, Ssample is the deviation standard of 5 67 67 1 0.75%
the sample, while tcritical is the t-test value with a certain level 6 67 69 2 2.90%
of confidence and degree of freedom. 7 67 66 2 2.29%
8 67 65 2 3.08%
S = Ssample / (n)1/2  9 67 65 3 3.88%
10 67 69 2 2.19%
Scomputed = tcritical × S  11 67 65 3 3.88%
12 67 64 4 5.51%
13 67 65 2 3.08%
TABLE III. SIMULATION RESULTS

Corridor Commercial Area TABLE V. PEDESTRIAN VOLUME IN THE CORRIDOR (2)


Simulation
Run- Speed Density Speed Density
(km/hour) (ped/m2) (km/hour) (ped/m2) To the Airport
1 2.801 0.0279 2.181 0.0133 Simulation Run- Volume Deviation
2 2.798 0.0287 2.248 0.0132 Observed Simulated Absolute %
3 2.698 0.0298 2.252 0.0131 1 63 63 0 0.00%
4 2.805 0.0282 2.248 0.0132 2 63 63 0 0.00%
5 2.745 0.0295 2.289 0.0130 3 63 63 0 0.00%
6 2.818 0.0292 2.325 0.0131 4 63 63 0 0.00%
7 2.819 0.0286 2.358 0.0125 5 63 63 0 0.00%
8 2.784 0.0287 2.276 0.0129 6 63 63 0 0.00%
9 2.831 0.0280 2.253 0.0132 7 63 63 0 0.00%
10 2.731 0.0297 2.226 0.0135 8 63 63 0 0.00%
11 2.820 0.0283 2.325 0.0126 9 63 63 0 0.00%
12 2.819 0.0291 2.318 0.0131 10 63 63 0 0.00%
13 2.938 0.0271 2.393 0.0123 11 63 63 0 0.00%
12 63 63 0 0.00%
By using a confidence level of 80%, the number of 13 63 63 0 0.00%
simulation repetitions needed to create the deviation standard

49
Advances in Engineering Research, volume 193

TABLE VI. PEDESTRIAN VOLUME IN COMMERCIAL AREA (1) distribution). The chi-square test aims to prove whether the
To the LRT Station observed frequency has a significant difference with the
Simulation Run- Volume Deviation expected frequency. The provisions for accepting hypotheses
Observed Simulated Absolute % in the chi-square test are as follows.
1 72 69 3 4.35%  If χcount2 ≤ χtable2, H0 will be accepted;
2 72 68 4 5.88%
3 72 71 2 2.13%  If χcount2 > χtable2, H0 will be rejected.
4 72 71 1 1.41%
5 72 69 4 5.11% TABLE VIII. THE GOODNESS OF FIT TEST (IN CORRIDOR)
6 72 73 1 1.37%
Corridor
7 72 71 2 2.13%
Simulation run- Speed
8 72 67 6 8.27% [(O-E)2]/E
Simulated Observed
9 72 67 6 8.27%
1 2.801 3.09 0.027
10 72 73 1 1.37%
2 2.798 3.09 0.0276
11 72 70 3 3.60%
3 2.698 3.09 0.0498
12 72 73 1 0.69%
4 2.805 3.09 0.0263
13 72 68 5 6.67%
5 2.745 3.09 0.0386
6 2.818 3.09 0.024
TABLE VII. PEDESTRIAN VOLUME IN COMMERCIAL AREA (2)
7 2.819 3.09 0.0237
To the Airport 8 2.784 3.09 0.0304
Simulation Run- Volume Deviation 9 2.831 3.09 0.0216
Observed Simulated Absolute % 10 2.731 3.09 0.0417
1 69 63 6 9.52% 11 2.820 3.09 0.0237
2 69 63 6 9.52% 12 2.819 3.09 0.0238
3 69 63 6 9.52% 13 2.938 3.09 0.0075
4 69 63 6 9.52% Average 2.800 3.09 Chi-Square
5 69 63 6 9.52% Deviation standard 0,058 0,000 0,3657
6 69 63 6 9.52%
7 69 63 6 9.52% TABLE IX. THE GOODNESS OF FIT TEST (IN COMMERCIAL AREA)
8 69 63 6 9.52%
Corridor
9 69 63 6 9.52%
Simulation run- Speed
10 69 63 6 9.52% [(O-E)2]/E
Simulated Observed
11 69 63 6 9.52%
1 2.181 3.025 0.2356
12 69 63 6 9.52%
2 2.248 3.025 0.1993
13 69 63 6 9.52%
3 2.252 3.025 0.1976
D. Model Validation 4 2.248 3.025 0.1994
Visualization is the most important thing when using a 5 2.289 3.025 0.1791
microscopic simulation model [9]. Therefore, Fig. 11 and 6 2.325 3.025 0.1619
Fig. 12 show the visualization of the micro-simulation 7 2.358 3.025 0.147
model, Viswalk, compared to the observation situation. 8 2.276 3.025 0.1853
9 2.253 3.025 0.1971
To ensure that walking behavior parameters have been
10 2.226 3.025 0.2113
properly calibrated and represent actual pedestrian behavior,
11 2.325 3.025 0.162
the simulation results are validated against the observed
12 2.318 3.025 0.1652
conditions. The parameter to be validated is the pedestrian
13 2.393 3.025 0.1322
speed in the corridor and commercial area on the sky bridge
Average 2.284 3.025 Chi-Square
of Sultan Mahmud Badaruddin II International Airport,
Deviation standard 0.058 0 2.3732
Palembang (see TABLE VIII and TABLE IX). The
validation method used is nonparametric statistical testing,
the goodness of fit test (following the chi-square

50
Advances in Engineering Research, volume 193

 A social (isotropic) : 1.3


 B social (isotropic) : 0.2
 VD : 6.0
 A social (mean) : 0.4
 B social (mean) : 2.8
 Noise : 1.2
 Queue order : 1.0
 Queue straight : 1.0
 Side preference : None

ACKNOWLEDGMENT
Most of the data and analysis contained in this paper have
been presented in the first author's final paper (unpublished)
in the Magister Program of Civil Engineering in Bandung
Institute of Technology under second author supervision.
More detailed work can be obtained in the university’s
electronic library.
Fig. 11. Visualization on Viswalk (corridor area).

REFERENCES
[1] H. Xiao, R. Ambadipudi, J. Hourdakis, and P. Michalopoulos,
“Methodology for Selecting Microscopic Simulators: Comparative
Evaluation of AIMSUN and VISSIM,” Retrieved from the University
of Minnesota Digital Conservancy, 2005.
[2] Austroads, “Guide to Project Evaluation. Austroads Incorporated,”
2006.
[3] S. Alexandersson, and E. Johansson, “Pedestrian in Microscopic
Traffic Simulation,” Gothenburg, Sweden: Chalmers University of
Technology, 2013.
[4] PTV, “PTV Vissim 8 User Manual,” Karlsruhe, Germany: PTV AG,
2015.
[5] M. Lagervall, and S. Samuelsson, “Microscopic Simulation of
Pedestrian Traffic in a Station Environment: A Study of Actual and
Desired Walking Speeds,” Sweden: Departement of Science and
Technology Linkoping University, 2014.
[6] D. Helbing, and P. Molnar, “Social Force Model for Pedestrian
Dynamics Physical Review,” 1995.
Fig. 12. Visualization on Viswalk (commercial area). [7] R. J. Galiza, I. Kim, L. Ferreira, and J. Laufer, “Modelling Pedestrian
CIrculation in Rail Transit Stations Using Micro-simulation,” 2011.
TABLE VIII and TABLE IX show the χcount2 values [8] FHWA, “Traffic Analysis Toolbox Volume III: Guidelines for
Applying Traffic Microsimulation Software,” 2004.
obtained are 0.3657 and 2.3732. On the other hand, the value
[9] B. Park, and J. D. Schneeberger, “Microscopic Simulation Model
of χtable2 derived from the t-test table is 26.22, which means Calibration and Validation: Case Study of VISSIM Simulation Model
that both χcount2 values are outside the critical area. Because for a Coordinated Actuated Signal System,” TRB 1856, pp. 185-192,
of χcount2 ≤ χtable2, then H0 is accepted and the microsimulation 2003.
model can represent the conditions of the observation.

V. CONCLUSIONS
Based on data analysis and discussion in the previous
section, it can be concluded that the Viswalk
microsimulation model of LRT station and sky bridge at
Sultan Mahmud Badaruddin II International Airport,
Palembang, has been statistically validated. The validation
test used is the goodness of fit (chi-square distribution) with
an average pedestrian speed indicator. Besides, the model
validation is also supported by the percentage deviations
between the number of pedestrians from observation and
simulation that have met the requirements. To achieve a
good level of model validity, the walking behavior
parameters on Viswalk are calibrated to the observational
conditions with the following results.
 Tau : 0.2
 React to n : 8.0
 Lambda : 0.3

51

You might also like