sustainability

Article
The Effect of Posted Speed Limit on the Dispersion of
Traffic Flow Speed
Chao Gao 1 , Jinliang Xu 1, *, Qunshan Li 2 and Jie Yang 1
1 School of Highway, Chang’an University, Xi’an 710064, China
2 Qinghai Province Traffic Construction Project Cost Management Station, Xining, Qinghai 810003, China
* Correspondence: xujinliang@chd.edu.cn




Received: 10 June 2019; Accepted: 26 June 2019; Published: 29 June 2019


Abstract: Speed dispersion is an important indicator to portray the quality of traffic flow and is
closely related to the road safety operation level. In order to clarify the influence of posted speed
limits on the dispersion of traffic flow speed, three sections with speed limits of 80 km/h, 100 km/h
and 120 km/h on the same expressway were selected for observation, and traffic volume, speed
and other parameters were collected. The characteristic speeds, such as average speed, V15 and
V85 , were evaluation indicators, where V15 and V85 are the speeds of the 15th and 85th percentiles
measured at the feature points of the road when the traffic is in a free-flow state and the weather is
good. The relationship between different posted speed limit values and the above indicators was
analyzed using the statistical analysis software, SPSS. The results show that the speed limit has a high
correlation with the average speed of traffic flow, V15 and V85 in free-flow state, with the coefficient
of determination being as high as 0.84, 0.85 and 0.92, respectively. In the restricted flow state, the
factors affecting the driver’s driving speed are mainly the decrease in driving freedom caused by the
increase of traffic volume rather than the speed limit value. In a free-flow state, when the posted
speed limit is increased and the average speed and the V85 also increased by approximately the same
magnitude. The posted speed limit values of 80 km/h, 100 km/h and 120 km/h correspond to the
90, 88 and 97 percentile speeds of the traffic flow, respectively. The higher the speed limit is, the
larger the speed difference between V15 and V85 becomes. The results of the study are very useful for
rationally determining the speed limit scheme under different traffic flows.

Keywords: speed dispersion; posted speed limit; traffic flow; expressway

1. Introduction
The discreteness of traffic flow speed can reflect the quality of traffic flow and is an important
concept in the field of traffic engineering. A number of factors have been found to increase the dispersion
of traffic flow speeds, including posted speed limits, intersections, on-street parking, pedestrian activity
and the composition of traffic flow. However, data constraints or insufficient samples prevent most
of the research conclusions on the subject from being widely accepted. China’s understanding of
the discreteness of traffic speed caused by speed limits is not comprehensive. Relevant researches
in developed countries mainly focus on the relationship between design speed, operating speed
and posted speed limit and have revealed some fully demonstrated results and conclusions that are
generally accepted. For example, relationship models between posted speed limit, operating speed,
design speed are established. Dixon et al. [1] selected multiple fixed observation sites in multi-lane
highways, increased the posted speed limit of these locations from 55 mph (88.6 km/h) to 65 mph
(104.7 km/h) and collected the velocity data before and after the posted speed limit change. The study
found that when the speed limit is increased, the speed of the free-flow state increases. From the data
collected by Parker [2] for an FHWA (Federal Highway Administration) study, it can be determined

Sustainability 2019, 11, 3594; doi:10.3390/su11133594 www.mdpi.com/journal/sustainability

Sustainability 2019, 11, 3594 2 of 15

that the average speed, the standard deviation of the speed and the 85th percentile speed differ by less
than 2 mph (3.2 km/h) before and after the speed limit change. Fitzpatrick surveyed 128 speed limit
intervals and found that the speeds of the 85th percentiles were all higher than the speed limit and the
speeds of the 50th percentiles were close to the speed limit [3].Fitzpatrick also pointed out that the 85th
speed should only be used as the benchmark for the speed limit. The actual speed limit should be
8~12 mile/h lower than the 85th speed [4] and based on the statistical data, a linear regression equation
of the operating speed and the posted speed limit is established [5]. Burritt [6], Dart et al. [7] found
that as the posted speed limit decreases, the average running speed also decreases; Ossiander et al [8],
Jernigan et al [9], Freedman and Esterlitz [10], Brow et al [11] and Upchurch [12] found that with a
national maximum speed limit from 55 mph to 65 mph, the average speed increased by 2–7 mph.
The above conclusions and results illustrate from different perspectives that speed limit changes
will have different effects on the speed of traffic flow. When the speed limit value changes by a certain
amount, the running speed will increase or decrease correspondingly in a certain proportion. At the
same time, the speed differences will also change as the speed limit value changes. Both speed and
speed difference are closely related to the degree of dispersion of the vehicle speed.
Vehicle speed dispersion is one of the concerns of speed research. Solomon (1964) [13] studies
the relationship between accident and vehicle speed, driver and vehicle and proposes the concept of
vehicle speed dispersion, which is defined as the difference in vehicle speed. Due to the differences in
research purposes and research methods, as well as the limitations of data collection and processing
methods, scholars have used a variety of vehicle speed distribution characteristics indicators to define
vehicle speed dispersion. In the current study, the distribution characteristics of vehicle speed mainly
include the sample mean value of the vehicle speed, the sample standard deviation, the sample
variance, the coefficient of variation (the ratio of the standard deviation to the mean), the skewness
and the peak value. According to the existing research results, it can be shown that the speed limit is
closely related to the degree of dispersion of the vehicle speed. Some scholars have collected the traffic
flow data before and after the adjustment of NMSL (National Maximum Speed Limit), attempted to
reveal the influence of speed limit on the dispersion degree of traffic flow speed through statistical
indicators using comparative analysis.Burritt [6], Forester et al [14], Rama [15] and other scholars
found that when the posted speed limit is reduced, the dispersion of the running speed distribution
decreases. Dart’s [7] research found that when the speed limit is reduced, the operating speed of
the road decreases and the dispersion of the speed distribution decreases, which is manifested by a
decrease in the standard deviation of the sample. Garber and Gadiraju [16] pointed out that the sample
variance of the running speed decreases with the increase of the speed limit. Some scholars believe that
after the speed limit standard is increased, the sample variance of the running speed has not changed
significantly. Kockelman [17] collected 1766 speed data from radar guns in Austin, studying the effects
of speed limits on speed and standard deviation of speed. This analysis demonstrated that the impact
of speed limits on vehicle speed variances is very small. It can be found that there are great differences,
even contradictions among the research conclusions.
Chinese scholars have also carried out a lot of research on the influence of speed limit on
vehicle speed. The researches mainly focus on the relationship between linear conditions, traffic flow
conditions, driving safety line of sight and speed limit [18–21]. Pei et al [18] determined speed limit
value of the curve segment and the slope segment through regression statistical analysis, in their
research, the data were acquired based on field data and passenger vehicle was selected as standard
vehicle form. Cheng Guozhu [19] analyzed the characteristics of expressway speed and established
a calculation model based on the maximum speed of the vehicle with the lowest generalized cost.
A mathematical method was used to obtain the optimal maximum speed limit reference value of
the expressway with different design speeds and different vehicle conditions. These research results
provide a certain theoretical accumulation for the study of speed limits in China. However, due to
the lack of systematic research on the data before and after the posted speed limit change, in order
to study the quantitative change of the characteristic velocity of the traffic flow, the research method
Sustainability 2019, 11, 3594 3 of 15

mostly adopts parallel alignment of different roads. Lijing Wang et al. [20] investigated the speed of
seven expressways and analyzed the distribution characteristics of the straight segments, slope section,
curved and slope section and curve section under different speed limit schemes, which proved that the
speed limit is an important factor affecting the running speed. In terms of the effect of speed limit on
vehicle speed dispersion, Wang Hao [21] conducted a comparative analysis of traffic flow data of three
different speed-limiting highways in China and Netherlands and used speed standard deviation and
speed variation coefficient to characterize vehicle speed dispersion. The speed standard deviation
increased with the increase of traffic flow speed and the speed variation coefficient decreased with the
increase of traffic flow speed.
Above all, scholars have studied the relationship between speed limit, design speed and operating
speed but the purpose of these studies is to use the relationship between them to design and construct
an expressway that meets speed expectations. There is still a lack of systematic and in-depth discussion
on the effects of vehicle speed dispersion. Although some studies have been conducted, there are
no convincing conclusions due to the small number and the consistency of the research samples, in
particular, only the velocity variance is used to characterize the speed dispersion, which cannot fully
explain the negative impact of vehicle speed distribution on traffic flow. Therefore, this paper intends
to select three sections on the expressway with the same vertical slope, flat curve radius and cross
section and collect traffic flow data at these three sections with posted speed limit of 80 km/h, 100 km/h
and 120 km/h. The vehicle speed and traffic volume data of the traffic flow using a sufficient sample
size under the speed value are analyzed, using statistical methods to discover the relationship between
the speed limit and the characteristic speed of the traffic flow and confirm if the influence of the speed
limit on the traffic flow quality is clearly determined.

2. Methods
In order to ensure the consistency of the observed samples, a field survey is used to establish the
relationship model between the speed limit and the dispersion of traffic flow speed. The survey needs
to eliminate the influence of road conditions. During the survey, it is necessary to control the survey
with the same traffic flow at each observation point.

2.1. Test Condition Control

The following basic requirements were met when the test sections and observation points are
arranged, in order to eliminate the influence of factors other than the posted speed limit on the
observation results:
Road sections such as long longitudinal slopes, sharp bends, construction areas, ramp entrances,
bridges and tunnel structures were avoided; two types of upper limit speed values were included
and the speed limit area was no less than 2 km; the speed limit sign was clearly visible, with no
occlusion; the position of the observation point was positioned 800m after the speed limit sign, which
was determined based on the short-term memory characteristics of the driver. The purpose of these
requirements was to ensure that the observation speed is the result of the "speed limit" effect; the
climatic conditions are ideal when observed.
In accordance to all the above requirements, the national expressway G65 straight section K900+000
~ K912 +000 was selected for observation. Figure 1 shows the details of each observation point.
 Sustainability 2019, 11, x FOR PEER REVIEW 4 of 15

Sustainability 2019, 11, 3594 4 of 15

Figure1.1.Distribution
Figure Distributionofofobservation
observationpoints.
points.
Figure 1. Distribution of observation points.
2.2.
2.2.Test
TestProcess
Process
The
Theessential
essentialofofthe
thefield
fieldexperiments
experimentsisistotoensure
ensurethatthatthe
thetraffic
trafficflow
flowcollected
collectedatatthe
thethree
three
observation points is basically the same. Therefore, the start and end of the traffic flow data
observation points is basically the same. Therefore, the start and end of the traffic flow data collection collection at
each observation 30
point are controlled by the same
at each observation point are controlled by characteristic vehicle. When
the same characteristic selecting
vehicle. a characteristic
When selecting a
vehicle
characteristic vehicle in the traffic flow, ensure that the vehicle is easily and quickly2discovered.
in the traffic flow, ensure that the vehicle is easily and quickly discovered. Figure shows the
basic flow
Figure of eachthe
2 shows 25basic
group of flow
trafficofflow
PSL tests.
each80 groupThrough theflow
of traffic above test Through
tests. process control, it is test
the above essentially
process
Relative mean error（%）

ensured that the vehicles at each

control, it is essentially ensured thatobservation
the point
vehicles are
at the
each same traffic
observation flow.
point are the same traffic flow.
PSL100
20
PSL120
15

10

0
0 0.2 0.4 0.6 0.8 1
v/C

Figure 5. Accuracy test of average speed prediction model.

Figure 2. Test flow chart.

Figure 2.Test flow chart.
The data collection time is from 8:30 am to 6:00 pm and the experiments time is from September
20, 2018 to October 7, 2018, for a total of 18 days, including working days and non-working days
and holidays. In each set of traffic flow surveys, the minimum number of vehicles required at each
Sustainability 2019, 11, 3594 5 of 15

observation point is greater than 320. After the data collection was completed, 150 sets of data were
obtained, totaling more than 228,000 vehicles.
At the same time, in order to avoid the influence of observers on driving behavior, laser-type
portable roadside traffic survey instruments were used at the observation points to collect cross-section
traffic flow data such as volume, speed, vehicle type and headway distance.

2.3. Data Preprocessing

In order to ensure the accuracy of survey data, each group of observation samples is subjected to
characteristic vehicle comparison checks, to ensure that the same traffic flow is collected by each of the
three observation points of each sample and the observation samples with inconsistent characteristic
vehicles are discarded. When all test group samples are reliable, SPSS was used to analyze the survey
data results. For the analysis, outlier data in the samples, such as the driver’s deceleration when
driving in the emergency lane and speed while drivers answered mobile phones, were discarded.
Finally, 105 sets of valid data were obtained, totaling more than 152,000 vehicles.

3. Results
In addition to the influence of posted limited speed (PSL), the vehicle running speed is also
affected by traffic volume. The close relationship between volume–velocity indicates that traffic volume
has an impact on traffic flow speed. Therefore, when analyzing the impact of speed limit on the vehicle
speed distribution, the traffic volume of the road must be considered. Based on the analysis of the
variation law of the characteristic speed with different posted speed limit under different traffic flow
patterns, the relationship between the posted speed limit and the characteristic speed is found out
and then the relationship model between them is established. The ratio v/C is used to describe the
flow patterns of expressway, it is the ratio of the maximum service traffic volume to the basic traffic
capacity. There are important differences in the design speed, number of lanes, traffic composition and
other factors of road sections. The “v” in v/C is an objective reflection of actual traffic flow, while “C”
accounts for modifications of the above factors. Its calculation formula is as follows [22]:
v
v/C = (1)
CB × fW × fHV × fP

where v is the traffic volume on the road, pcu/(h·ln); CB is the capacity volume of the road, pcu/(h·ln);
fW , fHV , fP are correction factors.
According to the Technical Standards for Highway Engineering of China [23], when v/C is less
than 0.55, the traffic flow belongs to free flow and when v/C is greater than 0.55, the traffic flow belongs
to the restricted flow. According to the field test data, 69 sets of data of free flow state (v/C less than
0.55) and 36 sets of data of non-free flow state (v/C greater than 0.55) were obtained.
The average sample speed of the vehicle speed, that is, the average speed, represents the overall
level of the traffic flow on the road segment. Figure 3 shows the average speed V T with the change of
v/C under different posted speed limit. The average speeds of traffic flow are all negatively correlated
with the degree of road congestion v/C and reveals that the slopes of the three trend lines are roughly
the same. The increase of the intercept is basically the same as the increase of the speed limit. For every
20km/h increase in the speed limit, the average speed is increased by about 18 km/h, which is generally
consistent with the increase of the speed limit value. It can be inferred that the average speed of the
traffic flow has a high correlation with the speed limit.
Sustainability 2019, 11, 3594 6 of 15
Sustainability 2019, 11, x FOR PEER REVIEW 6 of 15

120

Average Speed of Traffic Flow（km/h）

100

80

60

40

20 PSL 80km/h PSL 100km/h PSL 120km/h

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
v/C

Figure
Figure 3.
3. The
The average
average speed
speed of
of traffic
traffic flow
flow varies with v/C.
varies with v/C.

In order to verify the hypothesis, the test data groups were segmented for regression analysis
and model verification. In general, approximately 70% of data is used for model developments and
the remaining 30% data are used to to validate
validate the the goodness
goodness of of the
the model.
model. Therefore, the v/C ratios
collected during the test are sorted from small to large and test groups with a multiple of 3 were used
for model validation and the remaining serial numbers were used for regression analysis. analysis. According
statistical principle,
to statistical principle,under
underthethepremise
premisethat thatthe
the confidence
confidence level
level is 95%,
is 95%, thethe sample
sample sizesize
cancan reach
reach the
the accuracy
accuracy requirement
requirement of regression
of regression analysisanalysis
when when it reaches
it reaches 43. Thus,43.70
Thus, 70test
sets of setsdata
of test
weredata
usedwere
for
used for regression
regression analysis
analysis and and the remaining
the remaining 35 sets of35 sets
test of were
data test data
usedwere used for
for model model validation.
validation.
The 7070 sets
setsofoftest
testdata
datav/C
v/Cused
usedforfor regression
regression analysis
analysis were
were divided
divided intointo groups
groups of 0.05
of 0.05 unitsunits
and
and the average running speed of each group was obtained. Figure 4 is
the average running speed of each group was obtained. Figure 4 is a scatter plot of the relationship a scatter plot of the
between the average
relationship betweenspeed V T and the
the average PSL V
speed value. It is found that there is a certain relationship between
T and the PSL value. It is found that there is a certain
the average speed of the traffic flow and
relationship between the average speed of the trafficthe PSL value,
flowtheand
relationship model
the PSL value, between
the the average
relationship model
speed
between andthethe PSL is established
average speed and the by PSL
a regression:
is established by a regression:

VTT ==0.942PSL
V − 9.0618RR
0.942 PS−L9.0618 22
== 0.8352
0.8352 (2)
(2)

where:Where: V T indicates
V T indicates the average
the average PSL indicates
speed,speed, the posted
PSL indicates speed limit
the posted speedvalue, 80≤km/h
80 km/h
limit value, PSL ≤≤
120
PSLkm/h.
≤ 120 km/h.
It can be seen from the Formula (2) that the PSL has a significant effect on V T : when the PSL value
is in the range of 80~120 km/h, the average speed changes linearly with the change of the PSL and the
fitting coefficient of the model up to 0.8352 and the average speed value is lower than the PSL.
Although the model has a fitting factor of about 0.84, the model is validated by using the remaining
35 sets of experimental data to further verify the accuracy of the model. The verification results are
shown in Figure 5. As shown in Figure 5, the relative errors of the models are all below 10% and
most of them are below 5% under the free flow state (v/C ≤ 0.55). Considering the different driver
characteristics, the accuracy of the model is high. In addition, when the v/C ratio is greater than 0.55,
the relative error of the model is greater than 10% and the highest error is close to 30%, which indicates
that the accuracy of the model under non-free flow is not high. The reason for the analysis is that the
traffic flow density is gradually increased and the driving freedom is greatly affected. At this time, the
factors affecting the driver’s driving speed change are mainly the changes of the external environment
rather than the speed limit value.
 Sustainability 2019, 11, 3594 7 of 15
Sustainability 2019, 11, x FOR PEER REVIEW 7 of 15

100

Average Speed of Traffic Flow（km/h） 90

80

70

60
60 80 100 120 140
Posted
Figure 1. Distribution Speed Limit（km/h）
of observation points.

Figure 4. The
Figure average
4. The speed
average of traffic
speed flow—posted
of traffic speed
flow—posted limit
speed (PSL)
limit relationship
(PSL) model.
relationship model.

It can be seen from the Formula (2) that the PSL has a significant effect on V T : when the PSL
30
value is in the range of 80~120 km/h, the average speed changes linearly with the change of the PSL
and the fitting coefficient of the model up to 0.8352 and the average speed value is lower than the
PSL. 25 PSL 80
Relative mean error（%）

Although the model has a fitting factor of about 0.84, the model is validated by using the
PSL100
20
remaining 35 sets of experimental data to further verify the accuracy of the model. The verification
results are shown in Figure 5. As shown PSL120
in Figure 5, the relative errors of the models are all below
10% and most of them are 15 below 5% under the free flow state (v/C ≤ 0.55).Considering the different
driver characteristics, the accuracy of the model is high. In addition, when the v/C ratio is greater
than 0.55, the relative error of the model is greater than 10% and the highest error is close to 30%,
10
which indicates that the accuracy of the model under non-free flow is not high. The reason for the
analysis is that the traffic flow density is gradually increased and the driving freedom is greatly
affected. At this time, the5 factors affecting the driver’s driving speed change are mainly the changes
of the external environment rather than the speed limit value.
0
0 0.2 0.4 0.6 0.8 1
30
v/C

25 Figure 5.PSL
Figure 5. 80 test
Accuracy
Accuracy test of
of average
average speed
speed prediction
prediction model.
model.
Relative mean error（%）

On the cumulative frequency PSL10distribution curve of the running speed, the running speed
20 0
corresponding to K% is called the Kth percentile speed. The slope of the cumulative frequency
distribution curve is generally abruptly changed around the 85th and 15th positions, indicating that
the proportion of 15vehicles exceeding the 85th and less than the 15th is very small. The 85% running
speed is generally used as a reference for formulating the design speed and the basis for determining
the maximum speed 10 limit and the 15% vehicle speed is used as the basis for determining the minimum
speed limit of the expressway. Generally, the vehicle speed of the 85th percentile of all types of cars in
the free traffic flow 5state is used as the characteristic index of the operating speed.
Figure 6 is the speed cumulative frequency curve under each PSL value. The 80 km/h, 100 km/h
and 120km/h respectively correspond to the 90, 88 and 97 division speeds of the traffic flow. The PSL
0
value becomes larger and 0 the characteristic
0.2 speeds
0.4 such as V15 , V50 and
0.6 0.8V85 are correspondingly
1 larger,
v/C
speed is generally used as a reference for formulating the design speed and the basis for determining
the maximum speed limit and the 15% vehicle speed is used as the basis for determining the
minimum speed limit of the expressway. Generally, the vehicle speed of the 85th percentile of all
types of cars in the free traffic flow state is used as the characteristic index of the operating speed.
Figure2019,
Sustainability 6 is11,
the3594
speed cumulative frequency curve under each PSL value. The 80 km/h, 100 km/h 8 of 15
and 120km/h respectively correspond to the 90, 88 and 97 division speeds of the traffic flow. The PSL
value becomes larger and the characteristic speeds such as V15, V50 and V85 are correspondingly larger,
which
which also indicates that there is a certain change relationship between the speed characteristic speed
and the PSL.
PSL.

100%
90%
80%
Cumulative Frequency

70%
60%
50%
40%
30%
20% PSL 80km/h
PSL 100km/h
10%
PSL 120km/h
0%
10 20 30 40 50 60 70 80 90 100110120130140150
Speed (km/h)

Figure 6.
Figure 6. The Cumulative Frequency Distribution Curve of Running Speed.

Similarly, according
according to
tothe
themethod
methodofofthethe previous
previous analysis,
analysis, thethe 69 sets
69 sets datadata obtained
obtained in theinfree-
the
free-flow
flow statesstates are grouped,
are grouped, 46 sets
46 sets of test
of test datadata
werewere
usedused for regression
for regression analysis.
analysis. TheThe study
study divided
divided the
the
v/C v/C
ratioratio by steps
by 0.05 0.05 steps to calculate
to calculate the V85the
andVV8515 and V15ofspeeds
speeds of each
each step. step.
Figure Figure
7 shows the7relationship
shows the
between V85 between
relationship and V15 and
V85 the
andPSL. the PSL.relationship
Established
V15 and Established models between
relationship V85,between
models V15 and PSL
V85 , values:
V15 and
PSL values:
V = 1.025 PSL − 6.7022 R2 = 0.922
V8585= 1.025PSL − 6.7022 R2 = 0.922
(3)
(3)
V15==0.684PSL
V15 0.684 PSL + 4.456 RR22 ==0.853
+ 4.456 0.853 (4)
(4)
Sustainability 2019, 11, x FOR PEER REVIEW 9 of 15
The results of this study differ from those of scholar Fitzpatrick. Fitzpatrick [5] proposed a model
for predicting the operating speed of all highways, with the operating speed regression equation
EV85 = 7.657 + 0.98 PSL.
130 It can be found that in developed countries such as the United States, the
operating speed is higher than the speed limit, while the operating speed in China is lower than the
120
speed limit and the difference between the two is about y10= km/h.
1.038x -This
6.903situation is due to the fact that
R² = 0.941
high speed is generally considered to have an important impact on the full utilization of highway
V15、V85（km/h）

110
traffic efficiency. However, China's speed limit value has a greater constraint on vehicle travel,
resulting in many highways failing to fully meet their potential functions.
100
y = 0.6773x + 6.4818
90
R² = 0.8625

80

70 V85 V15

60
60 70 80 90 100 110 120 130
Posted Speed Limit（km/h）

Figure7.7. Scatter
Figure Scatter plot
plot of
of V
V85
85,, V
V15 and PSL.
15 and PSL.

To further verify the accuracy of the model, models (3) and (4) were validated using the
remaining 23 sets of data. The verification results are shown in Figure 8. It can be clearly seen in
Figure 8 that under the free flow state (v/C ≤ 0.55), the relative error value of the model is below 10%.
The accuracy of the model is relatively high considering the dispersion of the vehicle speed caused
by different driver characteristics.
Sustainability 2019, 11, 3594 9 of 15

The results of this study differ from those of scholar Fitzpatrick. Fitzpatrick [5] proposed a model
for predicting the operating speed of all highways, with the operating speed regression equation
EV85 = 7.657 + 0.98 PSL. It can be found that in developed countries such as the United States, the
operating speed is higher than the speed limit, while the operating speed in China is lower than the
speed limit and the difference between the two is about 10 km/h. This situation is due to the fact that
high speed is generally considered to have an important impact on the full utilization of highway traffic
efficiency. However, China’s speed limit value has a greater constraint on vehicle travel, resulting in
many highways failing to fully meet their potential functions.
To further verify the accuracy of the model, models (3) and (4) were validated using the remaining
23 sets of data. The verification results are shown in Figure 8. It can be clearly seen in Figure 8 that
under the free flow state (v/C ≤ 0.55), the relative error value of the model is below 10%. The accuracy
of the model is relatively high considering the dispersion of the vehicle speed caused by different
driver characteristics.

20.0

15.0 PSL 80
Relative mean error（%）

PSL 100
PSL 120
10.0

5.0

0.0
0 0.1 0.2 0.3 0.4 0.5 0.6
v/C

Figure
Figure 8.
8. Accuracy
Accuracy test
test of
of operating
operating speed
speed prediction
prediction model.
model.

In statistics, Range is used to describe the degree of dispersion of a set of data. Range is the
difference between the maximum and minimum values of a set of measured values. The larger the
range is, the greater the degree of dispersion, conversely, the lesser the degree of dispersion for a
smaller range. However, the contingency of the maximum and minimum values in the survey of
vehicle speed is too large, so the extreme difference can not reflect the dispersion of vehicle speed very
well. There are two commonly used parameters describing the speed distribution, the 85th percentile
speed and the 15th percentile speed. Replacing the range with the difference between V85 and V15
indicates the degree of velocity dispersion, which is more statistically significant, and the error is
smaller. If the value |V85 − V15 | is larger, it indicates that the more the vehicle speed changes between
vehicles, the vehicle speed distribution characteristics can be reflected to some extent.
Figure 9 is a scatter plot of the relationship between the Range and the PSL. It can be seen that, as
the PSL increases, the speed dispersion shows an increasing trend and the greater the PSL, the larger
the value of |V85 − V15 |.At the same time, it can be seen from Figure 4 that the average speed of traffic
flow becomes larger and the speeds of V15 and V85 are both increasing but the rising trend of V85 is
greater than V15 , which leads to an increase in speed dispersion. The main factors causing this gap are
vehicles with speed less than V15 and vehicles with speed greater than V85 . The "Technical Standard of
Highway Engineering" of China stipulates the dimensions of the design vehicle. In the expressway,
Sustainability 2019, 11, 3594 10 of 15

all models larger than the small passenger car are identified as large vehicles [24]. By analyzing the
proportion of large vehicles in the interval less than V15 and the proportion in the interval greater
than V85 , as show in Figure 10, it can be seen that the large vehicle is the largest ratio in the interval
of less than V15 . It is the smallest in the interval where the speed is greater than V85 and thus it can
be seen that the proportion of the large vehicle gradually decreases as the average speed increases.
Or it can be said that in the speed range less than V15 , V15 is mainly reduced due to the increase
of the proportion of large vehicles; in the speed range greater than V85 , V85 is increased due to the
increase of the over-speed ratio of the passenger cars, thus, the value of |V85 − V15 | is increased and the
Sustainability 2019, 11, x FOR
degree of dispersion ofPEER REVIEW
speed is increased. Therefore, it can be concluded that the two main 10 of 15
causes
of speed dispersion are vehicles over-speed and large vehicle ratio. Therefore, in order to reduce the
and V15 indicates the degree of velocity dispersion, which is more statistically significant, and the
speed dispersion of road vehicles and increase the road safety factor, specific measures can be taken
error
from isthesmaller.
aspectsIfofthe value V85the
improving − Vover-speed
15 is larger,problem
it indicates
andthat the more
reducing the the vehicle
mixed speed
traffic changes
of large and
small vehicles.
between vehicles, the vehicle speed distribution characteristics can be reflected to some extent.

35

30

25
V85－V15（km/h）

20

15

10

0
60 80 100 120 140
Posted Speed Limit（km/h）

Figure 9.
9. Scatter
Scatter plot
plot of
of Range
Range (V
(V85-V 15) and PSL.
Figure 85 -V15 ) and PSL.

Figure 9 is a the
In addition, scatter plot of
average the relationship
speed between
and the vehicle speedthe Range
range, theand the PSL.
vehicle It sample
speed can be seen
variance,
that, asstandard
the sample the PSL increases, the speed
deviation (SD), dispersionofshows
the coefficient an increasing
variation trend
(CV) and the and the
kurtosis greater
also affect and
the PSL, the larger the value of V85 − V15 .At the same time, it can be seen from Figure 4 that
determine the road traffic safety performance. Both the vehicle speed sample variance and the sample
standard deviation are measures of the magnitude of the vehicle speed sample fluctuations, reflecting
the average speed of traffic flow becomes larger and the speeds of V15 and V85 are both
the degree of deviation from the average vehicle speed. The larger the sample variance or sample
increasing but the rising trend of V85 is greater than V15, which leads to an increase in speed
standard deviation, the greater the fluctuation of the sample data, indicating that the vehicle speed is
dispersion. The main factors causing this gap are vehicles with speed less than V15 and
more discrete.
vehicles with speed greater than V85. The "Technical Standard of Highway Engineering" of
Figure 11 is the velocity-frequency curve of different speed limits in the free-flow patterns obtained
China stipulates the dimensions of the design vehicle. In the expressway, all models larger
through SPSS analysis. It can be found that all of them are normally distributed. With the increase of
than the small passenger car are identified as large vehicles [24]. By analyzing the
the PSL value, the speed of the traffic flow becomes larger; the top of the speed distribution curve at
proportion of large vehicles in the interval less than V15 and the proportion in the interval
PSL of 80 km/h is sharp and the kurtosis is large. The top of the distribution curve at PSL 100 km/h
greater than V85, as show in Figure 10, it can be seen that the large vehicle is the largest ratio
and 120 km/h gradually becomes lower, the speed limit value increases and the kurtosis of the speed
in the interval of less than V15. It is the smallest in the interval where the speed is greater
distribution curve decreases. It shows that the speed limit value increases, the speed distribution range
than V85 and thus it can be seen that the proportion of the large vehicle gradually decreases
becomes wider and the speed becomes discrete.
as the average speed increases. Or it can be said that in the speed range less than V15, V15 is
mainly reduced due to the increase of the proportion of large vehicles; in the speed range
greater than V85, V85 is increased due to the increase of the over-speed ratio of the passenger
cars, thus, the value of V85 − V15 is increased and the degree of dispersion of speed is
increased. Therefore, it can be concluded that the two main causes of speed dispersion are
vehicles over-speed and large vehicle ratio. Therefore, in order to reduce the speed
dispersion of road vehicles and increase the road safety factor, specific measures can be
taken from the aspects of improving the over-speed problem and reducing the mixed traffic
 50

Large vehicle r
40

30

Sustainability 2019, 11, 3594

20 11 of 15
10PEER REVIEW
Sustainability 2019, 11, x FOR 11 of 15

0
90 80 100 120
Proportion of large vehicles Less than V15
Posted Speed Limit（km/h）
80 Proportion of large vehicles in all vehicles
Figure 10.The proportion of the large vehicles in different speed ranges.
70 Proportion of large vehicles larger than V85

Large vehicle ratio（%）

In addition, the average speed and the vehicle speed range, the vehicle speed sample variance,
60
the sample standard deviation (SD), the coefficient of variation (CV) and the kurtosis also affect and
50 safety performance. Both the vehicle speed sample variance and the sample
determine the road traffic
standard deviation are measures of the magnitude of the vehicle speed sample fluctuations, reflecting
40
the degree of deviation from the average vehicle speed. The larger the sample variance or sample
standard deviation, the30greater the fluctuation of the sample data, indicating that the vehicle speed is
more discrete.
20
Figure 11 is the velocity-frequency curve of different speed limits in the free-flow patterns
obtained through SPSS 10analysis. It can be found that all of them are normally distributed. With the
increase of the PSL value, the speed of the traffic flow becomes larger; the top of the speed distribution
0
curve at PSL of 80 km/h is sharp and 80the kurtosis is large.
100 The top of the distribution
120 curve at PSL 100
km/h and 120 km/h gradually becomesPosted lower, Speed
the speed limit value increases and the kurtosis of the
Limit（km/h）
speed distribution curve decreases. It shows that the speed limit value increases, the speed
Figure
10. 10.The proportionof
Figure
distribution range Thewider
becomes and theofspeed
proportion the large
the large vehicles
vehicles
becomes inin
different speed
different
discrete. ranges.
speed ranges.

In addition, the average speed and the vehicle speed range, the vehicle speed sample variance,
the sample standard45deviation (SD), the coefficient of variation (CV) and the kurtosis also affect and
determine the road traffic safety performance. Both the vehicle speed sample variance and the sample
40
standard deviation are measures of the magnitude of the vehicle speedPSL 80km/h
sample fluctuations, reflecting
35 from the average vehicle speed. The larger the sample variance or sample
the degree of deviation
The proportion（%）

PSL 100km/h
standard deviation,30the greater the fluctuation of the sample data, indicating that the vehicle speed is
more discrete. PSL 120km/h
25 velocity-frequency curve of different speed limits in the free-flow patterns
Figure 11 is the
20 analysis. It can be found that all of them are normally distributed. With the
obtained through SPSS
increase of the PSL value, the speed of the traffic flow becomes larger; the top of the speed distribution
15
curve at PSL of 80 km/h is sharp and the kurtosis is large. The top of the distribution curve at PSL 100
km/h and 120 km/h10 gradually becomes lower, the speed limit value increases and the kurtosis of the
speed distribution curve
5 decreases. It shows that the speed limit value increases, the speed
distribution range becomes wider and the speed becomes discrete.
0
0 10 20 30 40 50 60 70 80 90 100110120130140150160
45 Posted Speed Limit（km/h）
40 PSL 80km/h
Figure 11. Velocity-Frequency curve.
35
The proportion（%）

PSL 100km/h
30
Standard Deviation (SD), which is most commonly used in probability statistics as a measure of
PSL 120km/h
25 standard deviation is defined as the arithmetic square root of the variance,
statistical dispersion. The
reflecting the degree of20dispersion between individuals within the group. Figure 12 shows that the
standard deviation (SD) of the average speed varies with v/C under different PSL. It can be clearly
15
reflected that the SD has different degrees of growth with the increase of the PSL. Compared with the
survey data with a PSL 10of 80 km/h, the standard deviation at a PSL of 100 km/h and 120 km/h increased
by about 5 km/h and 7.5 5km/h, respectively. On roads with higher posted speed limits, the SD of traffic
flow speed is larger, which
0 indicates that different speed limits have obvious influence on the discrete
0 10On
characteristics of vehicle speed. 20the
30 expressway
40 50 60 70where
80 90the
100110120130140150160
PSL is high, the expected vehicle speed
distribution range of different vehicles is larger than
Posted thatLimit（km/h）
Speed of the road section with a lower speed limit,
so the degree of speed dispersion is also affected. Under the same PSL, with the increase of traffic
volume, the dispersion degree of traffic flow has a slight downward trend. Analyzing the reason for
this downward trend it was found that, due to the traffic volume, the vehicle operation was restricted
by other vehicles in the traffic flow and the speed and freedom of driving were limited, so the speed
dispersion was also affected.
expected vehicle speed distribution range of different vehicles is larger than that of the road section
with a lower speed limit, so the degree of speed dispersion is also affected. Under the same PSL, with
the increase of traffic volume, the dispersion degree of traffic flow has a slight downward trend.
Analyzing the reason for this downward trend it was found that, due to the traffic volume, the vehicle
operation was
Sustainability 2019, restricted
11, 3594 by other vehicles in the traffic flow and the speed and freedom of driving
12 of 15
were limited, so the speed dispersion was also affected.

20
18

The standard deviation (SD) of the

16
average speed （km/h） 14
12
10
8
6
4
2 PSL 80km/h PSL 100km/h
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
v/C

Figure 12. The

Figure 12. The standard
standard deviation
deviation of
of the
the average
average speed varies with
speed varies with v/C.
v/C.

In order to make the comparison of the speed dispersion of traffic flow under different average
speeds more meaningful, the coefficient of variation (CV) is introduced, it is a relative description of
dispersion. The coefficient of variation (CV) is the ratio of the standard deviation of the
vehicle speed dispersion.
raw data to the average of the original data. Figure 13 depicts the coefficient of variation (CV) for the
different speed
average speed varies with v/C for different speed limits.
limits. It can be found that when the posted speed
limit is promoted, the CV is increasing but the effect of of traffic
traffic volume
volume isis not
not obvious.
obvious. The CV for the
limit speed of 80 km/h, 100 km/h and 120 km/h is approximately equal
speed of 80 km/h, 100 km/h and 120 km/h is approximately equal to 0.10, to 0.10, 0.130.13
andand
0.150.15
times the
times
average
the speed.
average ThisThis
speed. is different from
is different fromthethe
research
researchresults
resultsofofLeong
Leong[24].
[24]. When
When Leong
Leong compiled a
Australia's 31 suburban stations, it was found
compilation of speed measurements for three years in Australia’s
that the CV of the vehicle speed was about 0.17, the values under different speed limit standards
obtained from the survey are are smaller
smaller than
than those
those in
in developed
developed countries.
countries.
Sustainability 2019, 11, x FOR PEER REVIEW 13 of 15

0.25
The Coefficient of variation (CV) of the

0.20
average speed

0.15

0.10

0.05
PSL80km/h PSL100km/h PSL120km/h

0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
v/C

Figure 13. The average speed standard deviation Coefficient of variation (CV) varies with
with v/C.
v/C.

According to the previous analysis, as the PSL increases, the average speed V T increases,
V85 − V15 and the SD, CV of the average speed became larger. It can be seen that the dispersion
degree of traffic flow speed has a close relationship with the PSL. At the same time, the size of the
vehicle speed distribution interval determines the average speed of the vehicle, which in turn affects
Sustainability 2019, 11, 3594 13 of 15

According to the previous analysis, as the PSL increases, the average speed V T increases, |V85 − V15 |
and the SD, CV of the average speed became larger. It can be seen that the dispersion degree of traffic
flow speed has a close relationship with the PSL. At the same time, the size of the vehicle speed
distribution interval determines the average speed of the vehicle, which in turn affects the dispersion
degree of the traffic flow speed. Therefore, it can be concluded that the main action to control the
dispersion of road speed is to control the distribution range of vehicle speed.

4. Conclusions
In this study, through field experiments, the speed dispersion of the same traffic flow with different
speed limits was analyzed. Indicators such as the average speed V T , the SD, CV of the average speed,
the kurtosis of speed frequency curve and the cumulative frequency distribution curve of vehicle speed
were selected.
Based on the results obtained, we can conclude that: (1) In free-flow state, the posted speed
limit and the average speed of traffic flow, V15 and V85 have a high correlation and the coefficient
of determination is as high as 0.84, 0.85 and 0.92, respectively. As the posted speed limit increases,
the average speed and V85 will increase by about the same magnitude, the increase of V15 speed will
be lower than this magnitude. In the restricted flow state, the factors affecting the driver’s driving
speed are mainly the decrease in driving freedom caused by the increase of traffic volume rather than
the speed limit value (2) In a free-flow state, the speed cumulative frequency curve under different
posted speed limit standards indicates that the speed limit values of 80 km/h, 100 km/h and 120 km/h,
respectively, correspond to the 90, 88 and 97 division speeds of the traffic flow. The higher the posted
speed limit is, the larger the speed difference between V15 and V85 . The main factor causing this gap is
the distribution ratio of the large vehicles and passenger cars in the speed distribution interval and
the over-speed ratio of passenger cars. (3) The velocity-frequency curve distribution under different
speed limits in the free-flow state shows that as the speed limit value increases, the traffic flow speed
becomes larger; the speed limit curve with low speed limit is sharp at the top and the kurtosis is large.
The speed curve of the high speed is lower at the top, the kurtosis is reduced. As the speed limit value
increased, the speed distribution range is widened and the speed becomes discrete. (4) As the posted
speed limit increases, the SD of the average speed increases in varying degrees. Compared with the
survey data with posted speed limit of 80 km/h, the SD of the posted speed limit of 100 km/h and
120 km/h increased by about 5 km/h and 7.5 km/h, respectively. Under the same posted speed limit,
with the increase of traffic volume, the SD slightly decreased. (5) With the increase of the posted speed
limit, the CV of the average speed also increases. The CV at the speed limit of 80 km/h, 100 km/h and
120 km/h are 0.10, 0.13 and 0.15, respectively. China’s CV of the average speed is lower than those in
developed countries. At the same time, studies have shown that the effect of traffic volume on the
CV is not obvious. (6) The degree of dispersion of traffic flow speed has a close relationship with the
posted speed limit. The main actions to control the dispersion of speed should start with controlling
the speed distribution intervals.
The obtained results are very beneficial for the reasonable determination of the speed limit scheme
under different traffic flow patterns. However, due to the limitations of road infrastructure conditions,
only the traffic flow parameters of the same traffic flow under the three speed limit standards of
80 km/h, 100 km/h and 120 km/h were collected, the influence of more different speed limit standards
on the distribution of traffic flow speed remains to be further studied. Therefore, we will continue
to supplement this research in the future and study the impact of different posted speed limits on
road safety.

Author Contributions: Conceptualization, J.X.; Data curation, C.G., J.X. and J.Y.; Formal analysis, C.G. and J.Y.;
Investigation, C.G.; Methodology, C.G.; Resources, Q.L.; Supervision, J.X.; Writing – original draft, C.G.; Writing –
review & editing, J.X.
Sustainability 2019, 11, 3594 14 of 15

Funding: This research was funded by Shandong Provincial Department of Transport Technology Plan (Grant
no.2017B50) and Key Science and Technology Projects in the Transport Industry of the Ministry of Transport
(Grant no. 2018-ZD1-001, 2018-MS1-001).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Dixon, M.P. Evaluation of the Impacts of Reducing Truck Speeds on Interstate Highways in IDAHO; National
Institute for Advanced Transportation Technology University of Idaho: Moscow, Idaho, 2000.
2. Parker, M.R. Synthesis of Speed Zoning Practices; Department of Transportation Report FHWA/RD-85/096;
Federal Highway Administration: Washington, DC, USA, 1985; p. 8790.
3. Fitzpatrick, K. Design Speed, Operating Speed and Posted Speed Relationships. ITE J. 2003, 67, 5259.
4. Fitzpatrick, K. Is 85th Percentile Speed Used to Set Speed Limits? In Proceedings of the ITE 2002Annual
Meeting and Exhibit, Philadelphia, PA, USA, 4 August 2002; p. 3240.
5. Fitzpatrick, K.; Fambro, D. Predicting Operating Speeds on Tangent Sections of Two-Lane Rural Highways.
Transp. Res. Rec. 2000, 1737, 50–57. [CrossRef]
6. Burritt, B.E. Analysis of the Relation of Accidents and the 88-km/h (55-mph) Speed Limit on Arizona
Highways. Transp. Res. Rec. 1976, 34–35.
7. Dart, O.K. Effects of the 88.5-km/h (55-mph) speed limit and its enforcement on traffic speeds and accidents.
Transp. Res. Rec. 1984, 643, 23–32.
8. Ossiander, E.M.; Cummings, P. Freeway speed limits and traffic fatalities in Washington State.
Accid. Anal. Prev. 2002, 34, 13–18. [CrossRef]
9. Lynn, C.W.; Jernigan, J.D. The Impact of the 65 mph Speed Limit on irginia’s Rural Interstate Highways Through
1990; Report VTRC 92-R12; Transportation Research Council: Charlottesville, VA, USA, 1992; pp. 213–219.
10. Freedman, M.; Esterlitz, J.R. Effects of the 65-mph speed limit on speeds in three states. Transp. Res. Rec.
1993, 52–61.
11. Brown, D.B.; Maghsoodloo, S.; McArdle, M.E. The safety impact of the 65 mph speed limit: A case study
using Alabama accident records. J. Saf. Res. 1993, 125–139. [CrossRef]
12. Upchurch, J. Arizona’s experience with the 65-mph speed limit. Transp. Res. Rec. 1989, 1–6.
13. Solomon, D. Accidents on Main Rural Highways Related to Speed, Drivers and Vehicle; Bureau of Public Roads:
Washington, DC, USA, 1964.
14. Forester, T.H.; McNown, R.F.; Singell, L.D. A cost-benefit analysis of the 55 mph speed limit. South. Econ. J.
1984, 50, 631–641. [CrossRef]
15. Rama, P. Effects of weather-controlled variable speed limits and warning signs on driver behavior.
Transp. Res. Rec. 1999, 53–59. [CrossRef]
16. Garber Nicholas, J.; Gadiraju, R. Impacts of differential speed limits on the speed of traffic and the rate of
accidents. Transp. Res. Rec. 1995, 17–22.
17. Kockelman, K.; Bottom, J.; Kweon, J. Safety Impacts and Other Implications of Raised Speed Limits on High-Speed
Roads; National Cooperative Highway Research Program (NCHRP) Report; The National Academies Press:
Washington, DC, USA, 2006; pp. 17–23.
18. Pei, Y.; Cheng, G. Research on Operation Speed and Speed Limit for Freeways in China. J. Harbin Inst. Technol.
2003, 35, 168–172.
19. Cheng, G.-Z. Research on the Method of Setting Speed Limits on Freeway and Urban Expressway; Harbin University
of Technology: Harbin, China, 2007.
20. Wang, L. Research on Relationship between Speed Limits and Operating Speed for Freeway; Beijing University of
Technology: Beijing, China, 2011.
21. Wang, H.; Wang, W.; Chen, X.; Chen, J.; Li, J. Experimental features and characteristics of speed dispersion in
urban freeway traffic. Transp. Res. Rec. 2007, 1999, 150–160. [CrossRef]
22. Transportation Research Board (TRB). Highway Capacity Manual; Special Report 87; National ResearchCouncil:
Washington, DC, USA, 2000.
Sustainability 2019, 11, 3594 15 of 15

23. Technical Standards for Highway Engineering (JTG B01-2014); China Standard Press: Beijing, China, 2014.
24. Leong, H.J.W. The distribution and trend of free speeds on two-lane rural highways in New South Wales.
In Proceedings of the 4th Australian Road Research Board (ARRB) Conference, Melbourne, Australia, 1968;
Volume 4, pp. 791–814. Available online: https://trid.trb.org/view.aspx?id=1209374 (accessed on 29 June 2019).

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).