Content uploaded by Jingqin Gao
Author content
All content in this area was uploaded by Jingqin Gao on Aug 08, 2017
Content may be subject to copyright.
Gao, Ozbay 1
MODELING DOUBLE PARKING IMPACTS ON URBAN STREET
1
2
Jingqin Gao, M.Sc. (Corresponding author)
3
Graduate Research Assistant, Department of Civil and Urban Engineering;
4
New York University (NYU)
5
Six MetroTech Center, 4th Floor,
6
Brooklyn, NY 11201, USA
7
Tel: (646) 717-3652
8
E-mail: jingqin.gao@nyu.edu
9
10
Kaan Ozbay, Ph.D.
11
Professor, Department of Civil and Urban Engineering;
12
Center for Urban Science and Progress (CUSP);
13
New York University (NYU)
14
One MetroTech Center, 19th Floor,
15
Brooklyn, NY 11201, USA
16
Tel: (646) 997-0552, Fax: (646) 997-0560
17
E-mail: kaan.ozbay@nyu.edu
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Word count: 4696 Texts + 4 Table + 7 Figures = 7446
38
Abstract: 250
39
Reference: 486
40
Submission Date: August 1st, 2015
41
Revision Date: November 15th, 2015
42
43
44
Paper submitted for Presentation and Publication in the
45
Transportation Research Board’s 95th Annual Meeting, Washington, D.C., 2016
46
Gao, Ozbay 2
Abstract
1
Double parking (DP) is one of the key contributors to traffic congestion on urban streets.
2
Double parking violations of commercial vehicles while they load and unload at delivery locations
3
with insufficient curbside space can have significant negative impact on traffic. Motivated by the
4
need to study such impact in urban cities, this paper utilizes parking violation records for New
5
York City along with field data collected using video recording, and adopts a comprehensive
6
modeling approach that combines available data with two types of models. The first is an
7
M/M/ queueing model used to estimate double parking effect on the average travel time. The
8
second is a micro-simulation model developed and calibrated to study individual and combined
9
effects of various explanatory variables. Both models account for different effects of general
10
vehicles and commercial trucks. Via case studies in Midtown Manhattan and Downtown Brooklyn
11
(New York, US), double parking activities and driver behaviors are investigated and used for
12
comparative analysis. The M/M/ queueing model has been empirically validated using field data
13
collected as part of this study. Comparison results show a good fit for uncongested traffic
14
conditions. Micro-simulation results indicate different impact levels for 21 scenarios in four
15
categories namely, travel demand, double parking locations, frequency, and durations. This study
16
can provide traffic agencies a potential approach to quantify the impact of double parking in a
17
large-scale network and insights into the management and alleviation of on-street parking
18
problems including incentives for encouraging off-hour deliveries and more effective enforcement
19
during peak hours.
20
21
Gao, Ozbay 3
INTRODUCTION
1
In urban areas, obstructions of traffic such as double parking, commercial vehicle deliveries,
2
pedestrian jaywalking, taxi pick-ups and drop-offs, are potential impediments to road capacity and
3
vehicular speed, and causes traffic delay and safety risks. New York City Department of Finance
4
(NYCDOF) defines “double parking” as “standing or parking a vehicle on the roadway side of a
5
vehicle already stopped, standing or parked at the curb” (1). It is mainly due to the lack of available
6
on-street parking spaces and sometimes makes the street impassable, especially in one-way single-
7
lane situations. According to New York City Department of Transportation (NYCDOT) parking
8
regulations, double parking of passenger vehicles is illegal at all times in New York City (NYC),
9
regardless of location, purpose or duration (2). In most cases, a double parked vehicle will block
10
part of the street and a bus or a bicycle lane, if one is present. Double parking can become a serious
11
problem especially in congested urban areas since it is both obstructive and irritating.
12
Most cities rely on fines when dealing with double parking (3). In 2014, double parking
13
violation had 502,082 ticketed cases and was ranked seventh among all the parking violations in
14
New York City according to NYCDOF records. In reality, the total number of double parking
15
violations is highly underestimated as most of such activities may not even be recorded or ticketed.
16
Notably, it becomes more complex in mixed-used or commercial zones while double-parked
17
commercial vehicle can lead to further reduction in capacity and increase in travel time (TT) and
18
delay.
19
There are several other strategies to reduce double parking of commercial vehicles,
20
especially during the peak hours. A recent study that focused on Off-Hour Delivery (OHD)
21
strategies emphasizes the importance of promoting these strategies to reduce congestion caused by
22
double parking delivery trucks. On the one hand, driving on an OHD route saves about
23
$9,000/year/OHD-tour in parking fines since it is easier for truck drivers to find legal parking
24
spaces near their delivery destination during the off-hours (4). On the other hand, increased
25
enforcement of parking fines for double parking during the regular hours could encourage more
26
carriers to participate in OHD (5).
27
One of the challenges in quantifying the effects of double parking is the lack of reliable
28
analytical approaches dealing with this important problem. There’s a need to identify a robust
29
analytical methodology so that a large number of sites can be studied efficiently to quantify the
30
impact of double parking as well as various long and short term parking enforcement and
31
management strategies in a large urban network such as New York City. The use of a very detailed
32
microscopic simulation tool for this purpose might not be feasible due to time and budget
33
constraints.
34
As an alternative approach, a stochastic queuing model proposed in Gursoy-Baykal et al.
35
(6) to estimate incident induced delays is adopted to estimate average delayed travel time, given
36
the amount and duration of double parking activities. This kind of queuing model applicable to
37
transportation systems, can be applied at the macroscopic level, consists of multiple road segments
38
modeled as ‘individual servers’ that provide service to vehicles that join the queue. It has a set of
39
simplifying assumptions about vehicle arrival distribution, service distribution, and number of
40
servers to derive equations to predict the queue lengths and waiting times. It can be magnitude of
41
order inexpensive and faster to develop and implement such models compared with microscopic
42
simulation models. However, accuracy of such macroscopic models for various spatial-temporal
43
characteristics of a highly complex urban transportation network has to be validated carefully. This
44
paper describes such two models and discusses pros and cons of each approach along with
45
suggestions for improving both models.
46
Gao, Ozbay 4
LITERATURE REVIEW
1
Various studies have been taken into consideration for on-street parking and traffic congestion. In
2
2009, Portilla et al. (7) extended a model developed by Baykal-Gürsoy et al. (6) to quantify the
3
influence of parking maneuvers and badly parked cars on average link journey times. Baykal-
4
Gürsoy et al.’s model (6) was a M/M/ queueing model subjects to random interruptions of
5
exponentially distributed durations that was originally developed to simulate the effects incidents
6
have on delays that she had proposed in a previous article (8). It describes a queuing system where
7
arrival and service processes of the vehicles are all Poisson processes with infinite number of
8
servers. In Portilla et al.’s study, badly parked cars are introduced as time events while parking
9
maneuvers are introduced as high-frequency short duration event. The results showed a 15%-199%
10
increase in average journey time and a 6%-55% reduction in capacity. Guo et al. (9) evaluated
11
influence factors of on-street parking by using a proportional hazard-based duration model.
12
Narrow lanes, frequent turnover rates, and parking occupancy show a negative effect on travel
13
time.
14
However, only a few studies focus on the occurrence of double parking. Recent efforts
15
have primarily concentrated on the price strategies or applying microsimulation models to explore
16
on-street parking impact on traffic. Millard-Ball et al. (10) evaluated the first two-year
17
performance of SFpark (11) – a demand-responsive rate adjustment parking system embarked by
18
San Francisco Municipal Transportation Agency. Rate changes have helped achieve the city’s
19
occupancy goal (60%~80%) and reduced cruising by 50%. Another pilot called “PARK Smart”
20
(12) was implemented in New York City by NYCDOT. Schaller et al. (13) summarized that
21
pricing can be effective in achieving the goals of commercial loading availability, increasing
22
turnover, and parking availability. Morillo and Campos (14) showed that the economic cost differs
23
depending on the DP location and the type of typology.
24
Besides pricing strategies and economic impact, Kladeftiras and Antoniou (15) conducted
25
a research project using microsimulation to estimate the impact of average speed, delay and
26
stopped time, as well as environmental impact caused by double parking. An auxiliary lane is
27
created in the model for one-lane roadways for the vehicles to overtake the double-parked vehicle.
28
Results showed that by eliminating double-parking activities, it can increase 10% - 15% in average
29
speed and reduce 15% and 20% in delay and stopped time. Researchers in Italy also studied illegal
30
parking impact in a high-density urban area in Palermo, Italy (16). Results showed that parking
31
durations over 5 minutes would have significant negative impacts.
32
To sum up, most studies in the literature only provide insights into policy and economic
33
aspects of double parking. Only a few micro-simulation models are used to simulate double
34
parking impacts, but these are mainly outside the US and usually shown to be time-consuming and
35
labor intensive for network-wide implementation. Therefore, there is a need to identify a more
36
efficient macroscopic approach such as the queueing based model described in Gursoy-Baykal et
37
al. (6) to study the impact of double parking activities. Moreover, such macroscopic methods have
38
to be validated using real-world data and/or microscopic simulations. The feasibility of using
39
microscopic simulation for modeling double parking under various street configurations has to be
40
established since none of the major simulation tools has a default functionality for modeling double
41
parking. This paper attempts to address both these needs by collecting and analyzing field data and
42
validating a queueing based macroscopic model, as well as demonstrating a Paramics based
43
microscopic simulation model extended through a special application program interface (API)
44
developed by the research team to realistically modeling of double parking on urban streets.
45
46
Gao, Ozbay 5
DATA DESCRIPTION
1
NYCDOF’s parking violation records in 2014 are summarized to investigate the frequency of
2
double parking activities in NYC. As shown in the following table (TABLE 1), double parking
3
ranked as the seventh highest parking violation in NYC, which represents 5.0% of the total
4
NYCDOF’s parking violation records in 2014. Since most of the double parking events are
5
difficult to record, this number can be assumed to be significantly less than the real number of such
6
violations.
7
8
TABLE 1 Top 10 NYC Parking Violations in 2014
9
Rank
Violation Code
Tickets
Percentage
Violation Description
1
21
1,337,993
13.4%
Stand or park during street cleaning
2
38
1,302,810
13.1%
Muni meter: failing to show a receipt
3
14
884,828
8.9%
General no standing
4
37
745,096
7.5%
Muni Meter: excess of allowed time
5
20
584,334
5.9%
General no parking
6
71
556,677
5.6%
No Inspection Sticker
7
46
502,082
5.0%
Double parking
8
7
482,051
4.8%
Not stop at red light
9
40
479,191
4.8%
Stand or park beside a fire hydrant
10
36
457,141
4.6%
Exceeding speed limit in school zone
10
An automatic geocoding program is developed by the research team based on Google
11
Geocoding API (17) to convert the original address information into geographic coordinates that
12
can be employed to geolocate each event on a map.
13
Violation records are also visualized onto streets and neighborhoods to identify DP
14
“hotspots”. The goal of the geocoding and visualization is to understand the spatial-temporal
15
nature of double parking violations in New York City.
16
FIGURE 1 shows the results of four-month data from July to October in 2014. On the
17
neighborhood level, the neighborhood of Upper East Side, Upper West Side, and Midtown
18
Manhattan have the highest double parking violation records for all type of vehicles. The
19
neighborhood of Upper East Side, Upper West Side, and Hudson Yards-Chelsea-Flatiron-Union
20
Square ranked top three of double parking violation records for commercial vehicles. Double
21
parking has more violation records in commercial districts or mixed commercial/residential
22
districts, while commercial DP has less records compare to total DP in boroughs other than
23
Manhattan that are mainly residential. It must also be noted that Police enforcement is a crucial
24
factor to the number of summons for passenger cars. However, this may not necessarily reduce the
25
occurrence of commercial vehicle double parking in large and highly congested urban areas such
26
as NYC, since their loading and unloading activities are often along a short distance at multiple
27
destinations with not many opportunities of legal parking during a given time frame.
28
On the street level, 58th Street and Livingston Street ranked the first in Manhattan and
29
Brooklyn. The highest number is up to 797 total violations and 420 commercial violations on a
30
single street block in July to October in 2014.
31
Gao, Ozbay 6
1
(a) Total DP by street (b) Total DP by neighborhood
2
3
(c) Commercial DP by street (d) Commercial DP by neighborhood
4
5
FIGURE 1 NYC double parking violation tickets, 2014 July to October.
6
Gao, Ozbay 7
Case Study Sites
1
After identifying the hotspots, two case study sites, where the occurrence of double parking is
2
found to be significant from the above analysis, were chosen specifically (FIGURE 2).
3
1. Manhattan: 58th Street between 8th Avenue and 9th Avenue (highest number of records)
4
2. Brooklyn: Jay Street between Myrtle Avenue and Metrotech Center (observation of high
5
DP occurrence)
6
7
8
FIGURE 2 Double parking activities in study sites.
9
10
Traffic movements were recorded in April, July and November in 2015 during AM (8-
11
9AM), Midday (12-1PM), and PM (5-6PM) time periods for both sites to obtain traffic flow, speed,
12
queue length, DP information, and driver behaviors. The following observations are found from
13
the recorded videos:
14
Double parked trucks have more impact than general cars (e.g. further reduction on the
15
effective lane width).
16
Manhattan site has a shorter duration of DP events compared to the Brooklyn site.
17
The passing speed is generally higher than the default Paramics incident passing speed
18
(10MPH) which indicates that double parked events should not be treated same as
19
incidents.
20
Speed reduction is subject to multiple factors such as the location of DP.
21
For both sites, AM and Midday have noticeable number of DP events (20-30/hour), while
22
PM has very few DP events (1-2/hour). The regular curbside parking is not fully
23
occupied and the turnover rate is high during PM period.
24
25
Double parking information includes location, arrival time, departure time, vehicle type
26
and duration. A sample count sheet is shown in TABLE 2. The location of the DP events is
27
according to the th ( regular curb-parked vehicle measured from the upstream end
28
of the link. So the “location” shown in TABLE 2 stands for the double parking location parallel to
29
the regular curb parking. AM period is used for both study sites to better distinguish the delay
30
caused by double parking from the delay caused by midblock jaywalking pedestrians, which
31
occurs a lot during Midday.
32
33
Gao, Ozbay 8
TABLE 2 Sample DP Events
1
Site
Location
Arrival Time
Departure Time
Duration
Vehicle Type
Brooklyn
5
8:41:18
8:41:52
0:00:34
Car
Brooklyn
5
8:46:45
8:49:15
0:02:30
Car
Brooklyn
6
8:50:11
8:52:08
0:01:57
Truck
Brooklyn
2
8:51:15
8:51:55
0:00:40
Car
Brooklyn
5
8:52:52
8:56:53
0:04:01
Car
Brooklyn
2
8:57:09
9:04:17
0:07:08
Car
Manhattan
4
8:02:30
8:02:53
0:00:23
Taxi
Manhattan
1
8:04:40
8:06:16
0:01:36
Car
Manhattan
1
8:34:20
8:35:10
0:00:50
Truck
Manhattan
5
8:45:10
8:46:25
0:01:15
Car
2
Traffic information is collected and summarized for every 15 minutes (TABLE 3). One
3
important observation for the Brooklyn site is that the traffic was experiencing severe downstream
4
blocking during the last 15 minutes of the observed AM hour.
5
6
TABLE 3 Traffic Information for Case Studies
7
Location
Time
Volume
HV%
# DP
Status
Avg.TT(s)
Standard deviation(s)
Brooklyn
8:00-8:15 AM
66
12.1%
6
DP
5.52
1.19
w/o DP
5.26
0.93
8:15-8:30 AM
58
19.0%
3
DP
5.47
0.98
w/o DP
5.07
0.91
8:30-8:45 AM
59
18.6%
4
DP
6.01
1.83
w/o DP
6.00
0.82
8:45-9:00 AM
65
12.3%
8
DP
14.03
15.49
w/o DP
4.61
0.02
Manhattan
8:00-8:15 AM
114
17.5%
8
DP
7.86
2.54
w/o DP
6.74
1.43
8:15-8:30 AM
108
13.0%
6
DP
8.09
1.27
w/o DP
7.77
0.12
8:30-8:45 AM
107
15.9%
6
DP
8.15
1.40
w/o DP
6.93
0.98
8:45-9:00 AM
127
29.9%
7
DP
8.57
0.63
w/o DP
6.74
1.14
8
METHODOLOGY
9
M/M/ queueing model
10
A queueing model assumes that the space occupied by an vehicle on the roadway link represents
11
one queueing “server”, which starts its service once a vehicle joins the link and carries the
12
“service” until the end of the link is reached (18). The notation of queueing models is classified by
13
arrival distribution, service distribution and the number of servers. A queueing model that has a
14
Markovian arrival rate, a Markov modulated service rates, and an infinite number of servers in the
15
system is called M/M/ queueing model. A Markov process defines discrete states and the
16
Gao, Ozbay 9
transition probability from one state to another. One of the key features is that it solely depends on
1
the current state. The M/M/ model is a stochastic process that has the following features:
2
Arrivals process is a Poisson process at rate λ.
3
Exponential service times with parameter μ. There are always sufficient servers in the
4
system that every vehicle is served immediately after arriving.
5
When a vehicle arrives or departs, the system moves to an adjacent state
6
When double parking occurs, it indicates the high occupancy of the curbside parking lane.
7
This is also observed in the video with a turnover rate of only one or two vehicles per hour for
8
regular curbside parking. Thus, the regular curbside parking impact is negligible in this case study.
9
Adopted from (6, 7, 18), the system is designed with two server states “Failure (F)” or “Normal
10
(N)” to represent the conditions with or without double parking events. Let denote the rate of DP
11
clearance time and the frequency of DP events. The system will be in state if there are
12
vehicles in the system that are interrupted by double-parked vehicles, while the system will be in
13
state if there are vehicles in the system that are travelling without interruption. The
14
depiction of the modified M/M/ queueing model on a two-lane link is shown in FIGURE 3. The
15
servers in the systems will work at a low service rate μ’ when a double parking event happens
16
(FIGURE 4).
17
18
FIGURE 3 Modified M/M/ queueing model on a two-lane link (based on (6)).
19
FIGURE 4 State transitions for M/M/ model with two server states (based on(18)).
20
21
One important result from Baykal-Gursoy and Xiao (8) showed that the expected number of
22
vehicles on the link can be represented as below when the M/M/∞ queueing system is experiencing
23
service interruptions:
24
2( ') ( )( ')
E(X) (1 )
( ) ( ' ')
ff
r f r f
(1)
25
26
Average travel time on the link is:
27
r
…
…
…
…
0, N
λ
1, N
µ
0, F
λ
1, F
µ’
r
f
f
λ
3, N
3µ
λ
3, F
3µ’
r
f
λ
2, N
2μ
λ
2, F
2µ’
r
f
4µ’
λ
λ
4µ
(i+1)µ’
λ
λ
(i+1)µ
λ
i, N
iµ
λ
i, F
iµ’
r
f
…
…
…
…
Gao, Ozbay 10
2
( ) 1 ( ') ( )( ')
W = (1 )
( ) ( ' ')
E X f f
r f r f
(2)
1
To reflect road segment characteristics, Equation (2) is reformatted by Portilla et al. (7) to compute
2
average travel time t:
3
'
(1 ) ( )( ')
[1 (1 )]
1'
( ' )
vv
ff v v
LvL
tv v v
vf fv
d d L
(3)
4
Where L=length of the link (mile); D==traffic demand (veh/h); f=frequency of double parking
5
(events/hour), d=1/r=average duration time of double parking (h); v==average speed without
6
double parking (mph); and v’==average speed with double parking (mph).
7
8
With the intention of accommodating commercial vehicle violations, a double parked commercial
9
vehicle is treated as two double parking events in this study. Moreover, double parking impact can
10
vary among different street segments due to traffic volume, number of lanes, gross leasable area
11
(GLA) of commercial properties and so on. To capture these effects, equation (3) is modified by
12
adjusting a correction factor
as follows:
13
'
(1 ) ( )( ')
[1 (1 )]
1'
( ' )
f
vv
ff v v
LvL
tC v v v
vf fv
d d L
(4)
14
And,
15
1
1 Real TT
Model TT
N
fi
CN
(5)
16
Where Real TT = field travel time (s), Model TT = queueing model travel time (s), N = number of
17
observations.
18
19
We propose a correction factor of
= 1.07 for Manhattan site and
=1.02 for Brooklyn site
20
based on the observed data. However, since the observed data is limited, we are not capable of
21
generalizing
for all possible influencing factors. This will be the focus of future study where
22
automated data collection methods can be deployed.
23
24
Microsimulation Model
25
A Paramics (19) micro-simulation (FIGURE 5) is developed for the Brooklyn site to
26
simulate complex conditions especially when downstream blocking happens. The model is
27
bounded by Jay Street to the West, Flatbush Avenue Extension to the East, Tillary Street to the
28
North, and Willoughby Street to the South. The particular interest of this study focuses on the road
29
segment between Myrtle Avenue and Metrotech Center on Jay Street where DP data is available.
30
It’s a two-way segment has one 10-ft travel lane, one 4-ft bike lane and one 10-ft curbside parking
31
lane in each direction and is referred as the reference link in this study.
32
Paramics is a high-performance software that model the movement and behavior of
33
individual vehicles (19). To simulate double parking activities, an “incident” file coded in C
34
programming language is used to specify where, when and how long the “double parking event”
35
should occur. Double parking is programmed as a special incident type that has the event duration,
36
Gao, Ozbay 11
lane specification and overspills. The occurrence can be defined as either random or fixed on each
1
link. In order to test the location impact of DP event, this case study uses fixed incident with
2
defined distance and time.
3
Since the incident feature does not work well for single-lane situations while the vehicle
4
behind the double-parked event will not move until the event is cleared, a narrow auxiliary lane is
5
used along with a customized API plug-in to allow vehicles to overtake the DP vehicle. The API
6
plug-in is developed to slow down vehicle on a certain distance of the road link to simulate driver
7
behaviors when passing a DP vehicle. The location, length, and speed of the slow-down area can
8
be customized. The micro-simulation model is calibrated based on field traffic counts, average link
9
travel time and observed behaviors collected in April, 2015. It is also validated with Novemeber’s
10
data in the same year to demonstrate reasonable prediction capability. Validation results shows a
11
GEH (20) of traffic volume less than 0.20 and a root mean square error (RMSE) of travel time less
12
than 1.20 for all time intervals.
13
14
15
FIGURE 5 Case study area in Downtown Brooklyn and Paramics network.
16
17
The Traffic Engineering Handbook (21), along with other publications (7, 22), have addressed the
18
effect of curb parking, mainly caused by the lack of a curb lane plus the parking activity next to a
19
moving traffic stream. To estimate the impact on the average travel time due to different factors
20
under DP condition, 21 different scenarios were designed into four categories:
21
Category 1: Location of Double Parking Events Scenarios
22
Scenario 1-3: Double parking events happen at the beginning of the block, midblock, and
23
the end of the block
24
Category 2: Demand Scenarios
25
Scenario 4-7: 110%, 120%, 130%, 140% of morning peak demand
26
Category 3: Frequency of Double Parking Events Scenarios
27
Scenario 8-11: Increase one, two, three, and four double parking event every 15 minutes
28
Category 4: Duration of Double Parking Events Scenarios
29
Scenario 12-21: Average duration of the events is increased by one minute in each scenario
30
(e.g. Scenario 12 has one-minute average duration, Scenario 21 has 10-minute average
31
duration).
32
Gao, Ozbay 12
RESULTS AND DISCUSSION
1
First, M/M/ queueing model is applied to estimate average travel time with a 15 minute
2
time interval. FIGURE 6 and TABLE 4 provide a summary of the results. The result implies that
3
the M/M/ queueing model has a percentage difference of less than 8% for both sites in terms of
4
average travel times compared with the actual field travel times when no downstream queueing
5
exists. While applying correction factors to the queueing model, the percentage difference is
6
further reduced to less than 4%. In this study, the correction factor is tied to site-specific conditions.
7
However, by introducing it, we intend to introduce the idea of a general factor that will help
8
modelers to capture the impact of factors on traffic as a result of double parking. Many parameters
9
such as traffic volume, number of lanes, GLA of commercial properties may have an effect on the
10
correction factor. Further study and data collections are needed to define it as a general function
11
in terms of all these different characteristics of city streets as part of ongoing efforts.
12
Average field travel time significantly increased due to the downstream blocking of the
13
Brooklyn site in the last 15 minutes of the observed hour (FIGURE 6(a)). Under this circumstance,
14
M/M/ queueing model is found to be incapable of capturing certain spillover conditions which
15
resulted in the difference between the field data and the queueing model output. The reason for
16
this, is that M/M/ queueing model assumes independence of different queues, as well as the
17
independence of the arrival of double parking events, and does not account for spatial queue or
18
queue capacity.
19
20
(a) Brooklyn Site (Congested) (b) Manhattan Site (Uncongested)
21
22
FIGURE 6 M/M/ queueing model results.
23
24
TABLE 4 Summary of Field, Queueing Model, and Microsimulation Result
25
Site
Time
Field TT
Model TT
Model TT with Cf
Microsimulation TT
Brooklyn
8:00-8:15 AM
5.41
5.27
5.39
5.54
8:15-8:30 AM
5.30
5.08
5.20
5.83
8:30-8:45 AM
6.01
6.00
6.14
6.21
8:45-9:00 AM
13.60
4.95
5.07
12.91
Manhattan
8:00-8:15 AM
7.32
6.77
7.21
-
8:15-8:30 AM
8.03
7.78
8.29
-
8:30-8:45 AM
7.38
6.98
7.43
-
8:45-9:00 AM
7.40
6.82
7.26
-
Gao, Ozbay 13
The micro-simulation model is then used to quantify complex characteristics of double
1
parking events such as DP location. Travel time results are shown in the following heat maps
2
(FIGURE 7) for the four scenarios described in the previous section.
3
The first category tested the impact due to different DP locations. DP events at the end of
4
the block and midblock have similar mild effects on the average TT. Events occurring at the
5
beginning of the block have a 5.15% higher impact in terms of the average TT increase over the
6
AM peak hour. This matches the field observation that once a vehicle is double-parked near the
7
beginning of the block (close to the upstream end of the road in the direction of traffic flow),
8
drivers are more likely to slow down. DP vehicle under this condition usually reduces the effective
9
lane width and causes potential conflicts with other activities at the intersection such as pedestrian
10
crossing.
11
The second category of simulation scenarios examined the impact on the travel time of
12
applying 110%, 120%, 130% and 140% of morning peak demand. Generally, higher demand
13
resulted in higher travel time, as expected. The increase of the average travel time is 7.3%, 18.5%
14
and 18.6% for 20%, 30% and 40% increase in the traffic demand, respectively, while 10% increase
15
in demand does not cause a significant change in average travel time.
16
The third category of simulation scenarios accounted for the travel time effect if the
17
frequency of DP event increased. The hourly average travel time increased by 3.1%, 13.6%,
18
20.5%, and 27.6% when there is an increase of DP event frequency by 1, 2, 3, and 4 vehicles per
19
15 minutes, respectively.
20
The fourth category of scenarios specifies various duration time for DP events from 1
21
minute to 10 minutes. As shown in FIGURE 7, the longer the event goes on, the greater the increase
22
in the average travel time. Average travel time can be more than two times higher in the worst-
23
case scenario where the average duration of DP event is 10 minutes.
24
25
26
FIGURE 7 Microsimulation results.
27
Gao, Ozbay 14
CONCLUSIONS AND FUTURE WORK
1
This study examined a macroscopic M/M/ queueing model and micro-simulation for
2
estimating average travel time in the presence of double parking activities. Under uncongested
3
traffic conditions without downstream blocking, applying the M/M/ queueing model produced
4
a good fit with the field data. The observations obtained from video recording were used to develop
5
and calibrate a microsimulation model for one of the two case study areas namely, Downtown
6
Brooklyn. Simulation output shows that the location, frequency, and duration of the double parking
7
event and overall traffic demand have a certain impact on the average travel time.
8
Overall, M/M/ queueing model is an effective approach to compute average travel time
9
for unsaturated traffic flows under double parking conditions. For a large urban network, it has the
10
potential to be implemented instead of micro-simulation models mainly due to its computational
11
efficiency, if it can be combined with an appropriate correction factor. Effect of commercial
12
vehicles is also estimated in the model where one double-parked truck is assumed to account for
13
two DP events. However, when downstream blocking happens, the traffic condition does not
14
completely depend on what is hapenning on the current road segment. Micro-simulation is a more
15
powerful tool than the M/M/ queueing model for evaluating such congested scenarios and can
16
be used to examine individual and combined effects of various explanatory variables.
17
Future research will focus on collecting more field data using automated image processing
18
techniques, investigating specific impacts of double-parked commercial vehicles, exploration of
19
the accurate estimation of the correction factor function, correction factor for various spatio-
20
temporal characteristics of the transportation network, and developing new APIs to improve the
21
micro-simulation model for more challenging geometric and traffic conditions.
22
23
ACKNOWLEDGEMENT
24
The authors appreciate the New York City Department of Finance for providing data
25
through NYC open data (nycopendata.socrata.com). The contents of this paper only reflect views
26
of the authors and do not necessarily reflect the official views or policies of any agencies. The
27
authors would also like to thank Krzysztof Lukasik, a former graduate student of Rutgers
28
intelligent transportation systems (RITS) laboratory for working in the development of an earlier
29
version of the Paramics API plug-in used in this study.
30
31
REFERENCES
32
1. NYCDOF, Official Compilation of Rules of the City of New York, Chapter 39,
33
www1.nyc.gov/site/finance/vehicles/services-violation-codes.page, Accessed Jul. 26, 2015.
34
2. NYCDOT, NYC Parking Regulations, www.nyc.gov/html/dot/html/motorist/parking-
35
regulations.shtml, Accessed Jul. 26, 2015.
36
3. De Cerreño, A., Dynamics of on-street parking in large central cities. In Transportation Research
37
Record: Journal of the Transportation Research Board, No. 1898, 2004. pp. 130-137.
38
4. Rensselaer Polytechnic Institute, New York City Department of Transportation, Rutgers University.
39
Integrative Freight Demand Management in the New York City Metropolitan Area: Implementation
40
Phase. Publication No. RITARS-11-H-RPI.USDOT, 2013.
41
5. Holguin-Veras, J., K. Ozbay, A. Kornhauser, A. Shorris, and S. Ukkusuri. Integrative Freight
42
Demand Management in the New York City Metropolitan Area. Publication No: DTOS59-07-H-
43
0002. USDOT, 2010.
44
6. Baykal-Gursoy, M., W. Xiao, Z. Duan, and K. Ozbay, Delay estimation for traffic flow interrupted
45
by incidents. In: Proceedings of the 86th Annual Transportation Research Conf, 2006.
46
Gao, Ozbay 15
7. Portilla, A., B. Orena, J. Berodia, and F. Diaz, Using M/M/infinity queueing model in on-street
1
parking maneuvres. Journal of Transportation Engineering 135(8), 2009.
2
8. Baykal-Gursoy, M. and W. Xiao, Stochastic decomposition in M/M/∞ queues with Markov
3
modulated service rates. Queueing Systems 48(1-2), 2004, pp. 75-88.
4
9. Guo, H., Z. Gao, X. Yang, X. Zhao, and W. Wang, Modeling travel time under the influence of on-
5
street parking. Journal of Transportation Engineering 138(2), 2011, pp. 229-235.
6
10. Millard-Ball, A., R.R. Weinberger, and R.C. Hampshire, Is the curb 80% full or 20% empty?
7
Assessing the impacts of San Francisco’s parking pricing experiment. Transportation Research Part
8
A: Policy and Practice 63, 2014, pp. 76-92.
9
11. SFMTA, SFPark, sfpark.org, Accessed Jul. 26, 2015.
10
12. NYCDOT, Park Smart, www.nyc.gov/html/dot/html/motorist/parksmart.shtml, Accessed July. 26,
11
2015.
12
13. Schaller, B., D. Stein, and M. Blakeley, Parking Pricing and Curbside Management in New York
13
City. In: Proceedings of the Transportation Research Board 90th Annual Meeting 2011.
14
14. Morillo Carbonell, C. and J.M. Campos Cacheda, On-street illegal parking costs in urban areas.
15
Procedia-Social and Behavioral Sciences 160, 2014, pp. 342-351.
16
15. Kladeftiras, M. and C. Antoniou, Simulation-based assessment of double-parking impacts on traffic
17
and environmental conditions. In Transportation Research Record: Journal of the Transportation
18
Research Board, No. 2390, 2013. pp. 121-130.
19
16. Galatioto, F. and M. Bell, Simulation of Illegal Double Parking: Quantifying the Traffic and
20
Pollutant Impacts. In: Proceedings of the IV International SIIV Congress, 2007. pp. 12.
21
17. Google, Geocoding API, developers.google.com/maps/documentation/geocoding, Accessed July
22
31st, 2015.
23
18. Baykal-Gursoy, M. and Z. Duan, M/M/C queues with Markov modulated service processes. In:
24
Proceedings of the Proceedings of the 1st international conference on Performance evaluation
25
methodolgies and tools, 2006. pp. 38.
26
19. Quadstone, The paramics manuals, version 6.6. 1. Quastone Paramics LTD, Edinburgh, Scotland,
27
UK, 2009.
28
20. Chu, L., H.X. Liu, J.-S. Oh, and W. Recker, A calibration procedure for microscopic traffic
29
simulation. In: Proceedings of the Intelligent Transportation Systems, 2003. pp. 1574-1579.
30
21. Pline, J.L., Traffic engineering handbook, Prentice Hall, 1992.
31
22. Box, P.C., Curb-parking problems: Overview. Journal of Transportation Engineering 130(1),
32
2004, pp. 1-5.
33
34