Sei sulla pagina 1di 22

Transportation Research Part A 92 (2016) 162183

Contents lists available at ScienceDirect

Transportation Research Part A


journal homepage: www.elsevier.com/locate/tra

Capacity and safety analysis of hard-shoulder running (HSR).


A motorway case study
Marco Guerrieri , Raffaele Mauro
Department of Civil, Environmental and Mechanical Engineering, University of Trento, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Operational motorway conditions can be improved by introducing traffic flow
Received 11 July 2016 management and control systems, such as ramp metering (RM), high occupancy vehicle
Accepted 2 August 2016 (HOV) lanes, real-time variable speed limits (VSLs), reversible lanes (RL), automated high-
Available online 17 August 2016
way systems (AHS) and hard-shoulder running (HSR). The effects of such devices need to
be examined in terms of capacity and safety. This paper examines the case study of the
Keywords: Italian motorway A22, which is supposed to be equipped with an HSR system implemented
Hard-shoulder running
along 128 km in order to reduce congestion with consequent improvement in levels of
Capacity
Safety
service (LOS). We studied the traffic processes (capacity, flow distribution between lanes,
HSM reliability, etc.) and estimated the expected capacity and safety conditions. These latter
Reliability were studied with the method provided by the Highway Safety Manual (HSM), as well
Variable speed limit as by undertaking sensitivity analyses to quantify the expected changes in crash frequency
at varying HSR activation hours (from 30 to 200 h) in a year. It has been observed that HSR
activation does not involve significant variations in the general safety conditions in the
presence of a considerable capacity increase up to 35%. Moreover, have been identified
the cases which require speed limit implementation (with VSLs system) in function of
the values of reliability / and velocity process V, and also suggested a speed limit sign
system.
2016 Elsevier Ltd. All rights reserved.

1. Introduction

The increase in highway transport demand causes numerous problems, such as deteriorated levels of service (LOS), con-
gestion (oversaturation), delays, queue formation, higher energy consumption and pollutant emissions, and inadequate
safety conditions (Li et al., 2014). In order to alleviate these problems, we can intervene in infrastructures (e.g. additional
lanes, geometric design improvement, etc.) or in traffic flow management and control. The latter are particularly advanta-
geous in that they are environmentally sustainable and relatively cheap (Li et al., 2014) since they exploit the existing phys-
ical infrastructures. The most well-known highway traffic control systems are truck policies, ramp metering (RM), high
occupancy vehicle (HOV) lanes, real-time variable speed limits (VSLs), reversible lanes (RL), automated highway systems
(AHS) and hard-shoulder running (HSR).
The ramp metering aims to control and regulate traffic flow in and out of highways in order to obtain the intended oper-
ating conditions, for example in terms of LOS (Ramp Management and Control, Handbook, 2006; Papageorgiou et al.,1991).
VSL systems can be implemented to ease traffic congestion (by homogenizing vehicle driving speeds) and to increase
safety (Elefteriadou, 2014) with consequent 517% reduction in crashes (Lee et al., 2006; Elefteriadou, 2014).

Corresponding author at: DICAM, University of Trento, via Mesiano, 77, Trento, Italy.
E-mail address: marco.guerrieri@tin.it (M. Guerrieri).

http://dx.doi.org/10.1016/j.tra.2016.08.003
0965-8564/ 2016 Elsevier Ltd. All rights reserved.
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 163

In AHS automated or semi-automated vehicles, driving singly or in platoons, in dedicated and properly equipped lanes,
expand the lane capacity up to 8000 veh/h (Ioannou et al., 1997). Other advantages are the reduction in fuel consumption
and a sheer drop in pollutant emission values (up to 27%). In AHS vehicles drive in platoons (composed of no more than 20
vehicles), with inter-vehicle distances of 12 m, with a 3060 m interval between them (Ioannou et al., 1997). Numerous
research projects (e.g. ARAMIS, PROMETHEUS, VITA II, SARTRE, HAVEit, etc.) have shown AHS benefits (Carbaugh et al.,
1998; Papacostas and Prevedouro, 2001; Michael et al., 1998; Hamouda et al., 2009).
Hard-shoulder running (HSR) is a scheme enabling hard shoulders to be temporarily open to ordinary traffic, together
with general-purpose lanes, during peak demand hours. Some studies have shown that HSR can be also used in urban con-
texts as a support of smart cities (Wang et al., 2016). Sometimes priced dynamic shoulder lanes (PDSL) can also be of help
(NCHRP, 2012).
France, Holland and Germany were the first to introduce HSR in highways (Lemke and Moritz, 2001; Mattheis, 2002;
Moritz and Wirtz, 2003). In Germany its use dates back to 1996 (BBC News World Edition, 2004; Lemke and Irzik, 2006);
today it is used on a 200 km-long highway network, with a 100 km/h speed limit. In Holland HSR, built since 2003, is nowa-
days implemented on around 1000 km highway network. In Great Britain it has been working since 2006, with a 50 mph
speed limit. In Italy it is implemented on A14 (23.6 km-long) and on the Mestre bypass motorway. In North America
HSR installations are numerous: in San Diego (California) since 2005 on Highway I-805/SR 52 (the system operates when
traffic slows to 30 mph or lower), in Delaware, in Florida since 2005 on Highways SR 826 and SR 836, in Georgia on Highway
GA 400 since 2005 (the system operates when traffic slows to 35 mph or less), in Maryland on Highway US 29 (with 55 mph
as maximum speed limit), and finally in New Jersey, Virginia and Washington. Experiences on Dutch highways have shown a
capacity increase of 722 per cent (Kuhn, 2010; Elefteriadou, 2014; Taale, 2006). In Germany capacity increases up to 25%,
reduction in travel time up to 20% (FHWA, 2010) and in air and sound pollution (Brinckerhoff, 2010) have been measured. In
general but especially in Europe (particularly in Germany, Holland, England), HSR has been combined with VSLs (whose
speed is signalled by variable message boards) (Elefteriadou, 2014).
When designing HSR we should keep in mind a lot of factors which can affect its safety (e.g. hard shoulder width (Gross et al.,
2009; HSM, 2010)); conflicts at access ramps; sight distance; conflicts with motorists pulling onto the shoulder for emergency
purposes; need for bus driver training; speed differential between the general purpose lanes and the shoulder; return merge
distance adequacy; debris hazards on shoulder; reduced bridge clearance; drainage; emergency refuge areas (Kuhn, 2010)).
This research focuses on the case study of the HSR plan for the A22 motorway (Brenner motorway). The operational
motorway conditions and forecast capacities and expected crash frequencies (on the basis of the current configuration) have
been analysed. To this end has been used the methodology in the Highway Safety Manual (AASHTO, 2010). The study also
involved a sensitivity analysis on the annual hours when HSR is assumed to be open to traffic.

2. Case study: implementing a hard-shoulder running (HSR) measure on A22

The Brenner motorway (A22) is part of the trans-European road network TEN-T (specifically the HelsinkiLa Valletta cor-
ridor (V), see Regulation (EU) N. 1316/2013). Therefore, the A22 is subject to the rules in the Directive 2008/96/CE of the
European Parliament and Council (obligation to perform the Road Safety Impact Assessment RSIA for road infrastructure
interventions or traffic management), and has then a crucial role among the European and Italian national transport
infrastructures.
Currently the Brenner motorway A22 is 313-km long and has 24 toll booths, 4 junctions with other motorways (A1 North,
A1 South, A4 East and A4 West) (Fig. 1). The road has two lanes (3.75-m wide each) in each direction and right-hand hard
shoulders, each 3.45 m-wide; the central reservation is 1.20 m-wide (see Figs. 1 and 2).
The study forecasts the HSR implementation from Egna (km 102+00) to Verona North (km 230+00) for 128 km in total;
the passing lanes are 6 (3 in each carriageway) in this configuration. To this end we assessed the operational conditions of
the infrastructure (traffic curves, capacity, critical density, speed levels etc.) and estimated the variation in accident fre-
quency by means of the HSM methodology (HSM, 2010).

3. Theory

3.1. Some traffic flow theories

This paragraph provides the fundamentals of the traffic flow theory underlying the performed analyses, whose results are
given in the following sections. For further details and in-depth analyses see May (1990), Elefteriadou (2014), and Mauro
(2015). The study on functionality was carried out with macroscopic flow models which describe traffic flows as liquid or
gas motion. Dynamic variables are quantities locally aggregated such as traffic density k(x, t), flow q(x, t), average speed
v(x, t), or its variance r2v(x, t). Since these aggregations are local, these quantities generally vary over space (x) and time (t).
The main equation formally representing this theory is the conservation equation (Gerlough and Huber, 1975; Khne
et al., 1992; Barcel, 2010):
@qx; t @kx; t
0 1
@x @t
164 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 1. A22 motorway layout.

Fig. 2. A22 motorway cross section.


M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 165

Equation (1) expresses the traffic conservation law and is known as conservation equation: it has the same form as in any
continuous fluid. In order to solve (1) needs a further equation (equation of state) among the previous main quantities. Such
an equation is given by the experimental relationship between speed v and density, v = v(k), (or alternatively q = q(k)). Some
of the most well-known speed/density models v = v(k) are given in Tables 1 and 2 (May and Keller, 1967; Wang et al., 2009).
Where: vf is the free-flow speed and kjam is the jam density.

3.2. Estimation of motorway reliability

Denoting with vt1, vt, vt+1 the speed of the vehicles passing in a segment at time instants t  1, t and t + 1, we define the
speed level vt as the average of speed values. Be at the deviation between vt and vt shift of the speed:
vt vt at 2
The difference vt  vt1 was experimentally verified (Ferrari, 1988; Ferrari, 1991) to be proportional to the deviation at1
occurred at instant t  1:
vt  vt1 k  at1 3
where k is a coefficient (0 6 k 6 1). Denoting with h = 1 k, from the previous relations (3) and (4) we obtain:
vt vt1 at  h  at1 4
In the equation (4) the relation between the level vt at instant t and the level vtk at instant t  k is shown to be (Ferrari,
1988):
X
k
vt vtk k  atj 5
j1

Relation (5) shows that the sequence of levels vt is the realization of a random process called random walk, produced by
the random variable kat. Moreover, by defining wt = vt  vt1, from (2) we obtain:
wt at  1  k  at1 at  h  at1 6
at, at1, . . . represent a sequence of independent and identically distributed random variables, with zero mean and finite vari-
ance r2. Expression (6) shows an integrated moving average (MA) process of the first order (ARIMA (0,1,1)) (Mauro, 2015).
When the capacity q remains constant during the observation period, random variations of the level vt correspond to ran-
dom variations of the density kt. If vt decreases up to the value v = v(kjam) = q/kjam, kjam being the jam density, the flow
becomes unstable. Therefore, instability is a random event, whose occurrence depends on the process of the speed level
vt . The probability that no instability will occur in a duration T defines the reliability / of the vehicle flow for period T.
The / value for a traffic flow along the passing lane of a carriageway can be estimated with the expression (Ferrari, 1988):
 8:82
Q
/ 1  19:80   T1:983  M2 7
10; 000

Table 1
Deterministic single-regime speed density models.

Single regime model Function Parameters


 
Greenshields model v vf  1  kjam
k
vf; kjam
Greenberg model v vm  log kjam
k
vm; kjam
kk 
Underwood model v vf  e 10 k 2 vf; k0
2 k
Drakes model v vf  e

jam
vf; kjam
 n12 
Drew model v vf  1  kjam k
vf; kjam
 n
Pipes-Munjal model v vf  1  kjam k
vf; kjam

Table 2
Deterministic multi-regime speed-density models.

Multi-regime model Free-flow regime Congested regime


163:9
k 
Edie v 54:9  e (if k 6 50) v 26:8  ln 162:5
k
(if k > 50)
Two regime model v 60:9  0:515  k (if k 6 65) v 40  0:265  k (if k > 65)
145:5
Modified Greenberg v 48 (if k 6 35) v 32  ln k (if k > 35)
166 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

where Q = flow rate constrained along the offside lane (veh/h); T = period of time (min); M = absolute value of b, angular
coefficient of relationship r2 = a + bln(k) (m2 km s2). The coefficient r2v is the standard deviation of speeds vt.
As previously shown (Ferrari, 1988; Mauro et al., 2013) the relation r2v -ln(k) is linear, and the value of the angular coef-
ficient of the line is the parameter M in (7). On the estimation modalities of coefficients k and /, see the studies by Ferrari
(1988) and Mauro et al. (2013).
From (7) we obtain the capacity value Q which, with a given probability /, can pass along the lane with no instability
phenomena during the period T (Ferrari, 1988):
 8:82
1
1/
Q 10; 000  8
19:80  T  M2
Taking a reliability value close to 1 (/ usually = 0.90) and T = 0.25, (15 min), the flow rate defined by Eq. (8) is the capacity
of the passing lane (Q = Cpass); it depends on the characteristics of the traffic flow, expressed by M, and road and environmen-
tal conditions.
Given a carriageway with i lanes (i = 1, 2, . . ., n), on each of them passing the flow Qi, its capacity Cc is given by the total
passing flow when instability occurs on one of the lanes, that is the capacity deduced from (8) is reached for assigned /, T, M:
 8:82
1
X
n1
1/
Cc 10; 000  2
Qi 9
19:80  T  M 11

4. The Brenner motorway capacity, speed and reliability

4.1. Traffic data

The operating conditions of the A22 were examined in previous studies (Mauro, 2003, 2005, 2007; Mauro et al., 2013).
Compared to them, the new researches show some variations in the flow conditions, mostly deriving from the effects of over-
taking prohibition for the following vehicles:

 vehicles with weight > 7.5 t, from hour 0:00 to 24:00, from 0 to 85 km;
 vehicles with weight > 12 t, from hour 6:00 to 22:00, from 85 to 313 km.

Therefore, in these time slots such vehicles must compulsorily travel on the right lane. The prominent segments chosen to
carry out the analyses are reported in Table 3 together with the surrounding geometric conditions, in terms of horizontal and
vertical alignments.
The examined traffic data were collected from 5th to 11th May 2014 and from 8th to 14th December 2014. For some seg-
ments (Adige, Rovereto) we further considered the data concerning the year 2003, utilized in previous studies (Mauro, 2005).

4.2. Inference of flow relationships

The macroscopic flow parameters - capacity q; speed v (harmonic mean); density k were calculated with reference to
time intervals of 5 and 15 min. The vehicles were homogenised by means of HCM (2010) coefficients (flows are then
expressed in uvp/h). The number of the pairs (v; k), (q; k), (v; q), set in intervals DT = 5 min, is shown in Table 4. The total
number is N5 = 24,192.
We first considered the main flow models v = v(k) for each right lane, passing lane and carriageway available in the sci-
entific literature (see Tables 1 and 2). We opted for the bell-shaped curve model proposed by Drake (May, 1990) which
proved to be the best in interpreting the available data. Drakes equation (see Table 1), after logarithmic transformations,
is as follows:
2
k
lnv lnvf  2
10
2  kjam

or
V1 a bD1 11

Table 3
Locations of observation sections and indications of surrounding geometric conditions.

Observation section Location Horizontal alignment Vertical alignment; slope (%)


Kofler 063+500 Tangent Curve: R = 10.000 m; slope = 0.41%
S. Michele 123+960 Tangent Tangent, slope = 0.03%
Portale Affi 205+500 Curve: R = 1000 m Curve: R = 10.000 m; slope = 0.23%
Mantova 271+900 Tangent Curve: R = 15.000 m; slope = 0.00%
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 167

Table 4
Numbers of the determination (v; k), (q; k), (v; q) - year 2014.

San Michele San Michele Kofler Kofler 063+500 Portale Affi Portale Affi Mantova Mantova
123+960 Nord 123+960 Sud 063+500 Nord Sud 205+500 Nord 205+500 Sud 271+900 Nord 271+900 Sud
4032 4032 4032 4032 2016 2016 2016 2016

In which V1 = ln(v); a = ln (vf); b = 1/(2k2jam); D1 = k2 where vf denotes the free-flow speed and kjam the critical density
(jam density), i.e. the density associated with the capacity. By means of the fundamental flow relation q = kv, we obtain
the following:
v
u vf
uln
q v  t 0:5v 12
k2jam

 2
12 k
q vf  k  e kjam
13
Which allow the two additional flow relations to be traced: v = v(q) and q = q(k). Finally, the flow models were calibrated.
By observing (10) we realized that the two parameters needed to be assessed: free-flow speed vf and critical density kjam. On
the basis of point scattering (k2; ln(v)) we estimated the line to the minimum square (11) and then determined the model
calibration parameters vf and kjam.
For example, the Fig. 3 shows the speeddensity scatter points (k2; ln(v)) for the right lane of the section Kofler (southbound
q q
roadway, cfr. Table 3); we obtain: vf = exp(a) = exp(4.3675) = 78.85 km/h; kjam = 2b 1 1
= 20:0005 31:62 pcu/km/lane.

Finally, we obtain: q 78:85  k  e231:62 .


1 k 2

By way of an example, Figs. 4 and 5 show the flow curves in San Michele (route segment), Highway South.
Tables 58 illustrate the flow relationship parameters for some of the examined sections (integrated with surveys made
in 2003).

4.3. Lane flow distribution

On every significant day in the year 2014 we determined the percentage frequency distribution of slow vehicles for lane
and lane flow distribution (Figs. 6 and 7).
Figs. 6 and 7 show that the flow on the passing lane increases at increasing flow on the carriageway Qt and, up to a certain
value of Qt (e.g. Qt = 1700 pcu/h for the case in Fig. 6) it is below the capacity on the right lane. Over Qt the situation is just
the opposite.
Figs. 811 show the typical histograms of the percentage frequency distribution of heavy vehicles on the total flow per
lane. Thus we verified that heavy vehicles utilize almost exclusively the right and not the passing lane, in accordance with
the overtaking prohibition imposed. A quantitative comparison between the distributions without (year 2003) and with
(year 2014) overtaking prohibitions is given in Table 9. The coefficient of variation was estimated as ratio between the square
deviation and the average. For the degree of homogeneity we assumed the statistics

Fig. 3. Speeddensity scatter points, right lane of the section Kofler (southbound roadway).
168 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 4. Speed - flow diagrams for the right lane (section san Michele, northbound roadway).

Fig. 5. Speed - flow diagrams for the right lane, passing lane and carriageway (section san Michele, southbound roadway).

Table 5
Traffic flow parameters (section San Michele).

Lane - carriageway San Michele (km 123+960) - northbound roadway San Michele (km 123+960) - southbound roadway
vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h] vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h]
Right lane 108 23 1484 66 109 21 1420 68
Passing lane 134 23 1901 81 129 22 1747 78
Carriageway 121 46 3361 73 110 50 3340 68

Table 6
Traffic flow parameters (section Mantova).

Lane - carriageway Mantova (km 271+900) - northbound roadway Mantova (km 271+900) - southbound roadway
vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h] vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h]
Right lane 103 32 1968 62
Passing lane 126 29 2209 76
Carriageway
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 169

Table 7
Traffic flow parameters (section Rovereto).

Lane - carriageway Rovereto (km 161+100) - northbound roadway Rovereto (km 161+100) - southbound roadway
vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h] vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h]
Right lane 105 23 1493 64
Passing lane 127 26 2012 77
Carriageway 114 50 3468 69

Table 8
Traffic flow parameters (section Adige).

Lane - carriageway Adige (187+300) - northbound roadway Adige (187+300) - southbound roadway
vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h] vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h]
Right lane 103 21 1341 63 109 24 1607 66
Passing lane 120 22 1581 73 129 26 2043 78
Carriageway 113 40 2703 68 116 51 3621 71

Fig. 6. Relationship between flow rate of lane and total flow rate Qt (San Michele, southbound roadway).

Fig. 7. Relationship between percentages of flow rate of lane and total flow rate Qt (San Michele, southbound roadway).
170 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 8. Frequency distribution of the percentage of heavy vehicle. Section San Michele, northbound roadway, right lane, weekdays.

Fig. 9. Frequency distribution of the percentage of heavy vehicle. Section San Michele, northbound roadway, passing lane, weekdays.

Fig. 10. Frequency distribution of the percentage of heavy vehicle. Section San Michele, northbound roadway, right lane, weekends.

Hq
g1 14
Hmax
where Hmax = ln (m), with m number of classes of the frequency distribution which recorded a non-zero percentage, and
P
Hq  m i1 qi lnqi (entropy), qi being the frequency values related to the classes of statistical distribution of the character
heavy vehicles percentage.

4.4. Speed regime (characteristics of the average speed regime)

The characterization of the difference between speed regimes in right lanes and passing lanes was obtained by analysing
the combined mean speed (mean of instantaneous speeds) and flow performances over time, in wide-ranging flow condi-
tions, from the free-flow to congestion. A typical speed process is given in Fig. 12.
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 171

Fig. 11. Frequency distribution of the percentage of heavy vehicle. Section San Michele, northbound roadway, passing lane, weekends.

Table 9
Comparison between the distributions without (year 2003) and with (year 2014) overtaking prohibitions.

Lane/days Mean Variance Coefficient of variation Degree of homogeneity


Right lane/weekdays (without overtaking prohibitions, year 2003) 55.36 564.28 0.43 0.14
Passing lane/weekdays (without overtaking prohibitions, year 2003) 13.56 464.75 1.64 0.69
Right lane/weekdays (with overtaking prohibitions, year 2014) 21.85 306.11 0.80 0.24
Passing lane/weekdays (with overtaking prohibitions, year 2014) 3.21 82.34 2.83 0.88

Fig. 12. Example of speed and flow processes. Section S. Michele, northbound roadway (weekday).

The analyses allowed to identify some perturbation phenomena in the traffic flow, with queue formation and stop-and-go
conditions, for example when right lanes are closed for road works or other reasons (e.g. accidents).
In Fig. 13, at section km 063+500 (Kofler near tunnel) we can observe forced or breakdown flow only on the right lane,
while in Fig. 14 there are clear saturation phenomena in both lanes. From the data processing can be noted:

 in stationary conditions, the speed performances for both lanes are basically always parallel to each other;
 in the passing lane speeds are nearly always higher than those in the right lane, also as a consequence of a no overtaking
restriction for heavy vehicles;
 even if vehicles are conditioned markedly and progressively, in the presence of steady flows, left-hand drivers adopt
higher speeds than those on the right-hand side;
 when congestion is attained more rapidly, the shift from transient to forced flow (and consequently, from a certain instant
in time, to the practical impossibility of distinguishing between the speed values in the two lanes) generally involves a
higher gradient for real-time speed performance in the passing lane than in the right lane;
 when leaving congestion or conditional flow towards steady or free-flowing conditions, have been noted, at the same
instant in time, higher speeds in the left than in the right lane and the former evolve more rapidly than the latter towards
higher values at the arrival state.
172 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 13. Speed processes (continuous line) and flow (dotted line). Section Kofler northbound roadway, right lane. Stop-and-go conditions on the right lane.

Fig. 14. Speed and flow processes. Section Kofler northbound roadway. Stop-and-go conditions on both lanes.

Table 10
Regression line and values of coefficient M. source: Mauro et al., 2013

Observation section (roadway) Regression line R2 M


S. Michele (southbound) r = 4.131  ln D + 15.540
2
0.487 4.131
S. Michele (northbound) r2 = 4.351  ln D + 16.319 0.789 4.351

By studying speed processes can be obtained the laws r2 = a + bln(k), the coefficient M = |b| and therefore the reliability
/ = /(Q, T, M) (see Eq. (8)). An example is given in Table 10 showing the values for the section San Michele (Mauro et al.,
2013).
Finally, we examined the performances of average vehicle headways sm (for 50 intervals). A typical performance is drawn
in Fig. 15. We can see that the average headways on the right lane are nearly always shorter than those in the passing lane,
which can be accounted for by the lower number of flows in the latter, at least up to carriage flows of around 1700 pcu/h
(Qt = 1700 pcu/h, see Figs. 6 and 7). It is worth noting that in the night hours the flows are modest, the average vehicle head-
ways are very high (sm = 1/Q) and drivers speeds often break speed limits imposed on the A22.

5. HSR: capacity and safety evaluations

The HSR capacity can be estimated by analogy with what occurs in motorways with three lanes on each carriageway.
More precisely, a driver opts for the HSR lane if there is a low probability that he will reach a slower vehicle, or a high
probability that, once reached, he will accept its speed, mainly without changing lanes. The central and left-hand lanes of
a three-lane carriageway are basically utilized in the same way as the middle and outside lanes in a two-lane carriageway,
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 173

Fig. 15. Headway and speed processes. Section Kofler northbound roadway.

Table 11
Estimated traffic flow parameters, A22 motorway with HSR (section San Michele).

Lane/carriageway San Michele (km 123+960) - northbound roadway San Michele (km 123+960) - southbound roadway
vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h] vf [km/h] kjam [pcu/lane/km] C [pcu/h] vjam [km/h]
Right lane (HSR) 82 23 1150 50 89 20 1080 54
Central lane 108 23 1484 66 109 21 1420 68
Passing lane 134 23 1901 81 129 22 1747 78
Carriageway 108 69 4535 66 111 63 4247 67

respectively. This working scheme of a three-lane carriageway is consistent with the results of traffic surveys made on high-
way segments with two or three lanes passed by traffic flow vehicles of a comparable layouts. For instance, the maximum
vehicle capacities in a three-lane A1 motorway section between Modena and Bologna were, respectively, from left to right,
1902 pcu/h; 1504 pcu/h; 967 pcu/h; the maximum flow volumes recorded in a two-lane A14 motorway carriageway
between Bologna and Imola, again from left to right, were 1889 pcu/h and 1465 pcu/h (Mauro, 2005).
Therefore, the capacities in the central and left-hand lanes of a three-lane carriageway can be virtually considered to coin-
cide with the right- and left-hand lanes (passing lane) of the two-lane carriageway. In such a hypothesis, the central and left-
hand lanes are given the determinations of the flow parameters in Tables 58.
The determinations of (vf)right and (kjam)right concerning the right-hand lane are obtained by imposing on the central lane
operating characteristics averaging those in the right- and left-hand lanes (otherwise indicated below as outer and inner lane
respectively). This means to solve, with regard to (vf)right and (kjam)right, the following two equations:
vf pass vf right
vf central 15
2

kjam pass kjam right


kjam central 16
2
 2
kjam right
12
Once (vf)right and (kjam)right are known, we obtain (vjam)right with the expression vjam right vf right  e kjam right , and

then the capacity C = (vjam)right  (kjam)right.


The carriageway capacity can be estimated as the sum of the capacities of the three lanes; a similar consideration
can be made for the critical density (kjam), while vf and vjam are deduced from the expressions vf = C/[kjam  exp(0.5)] and
vjam = C/kjam.
By way of an example Table 11 shows the flow parameter values obtained for the section San Michele (data 2014). For the
same section Fig. 16 shows the relations v = v(q) and Fig. 17 the relations v = v(k). For the northbound roadway the sum of
the lane capacities without HSR is 3361 pcu/h while it adds to 4535 pcu/h (35% increase) with active HSR; on the other hand,
for the southbound roadway it grows from 3340 pcu/h to 4247 pcu/h (27% increase) (see Table 11).
We then focused on the safety effects deriving from HSR activation from Egna (km 102+00) to Verona North (km 230+00),
necessary to achieve the capacity increases discussed above.
174 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 16. Speed - flow diagrams for the lanes and carriageway (section san Michele, northbound roadway).

Fig. 17. v = v(k) for the lanes and carriageway (section san Michele, southbound roadway).

In order to estimate safety conditions we first carried out the following analyses:

1. A22 subdivision into 25 basic segments (see Table 12) with constant AADT;
2. detailed study of the geometry of each basic segment (e.g. curvatures);
3. geometric examination of the A22 configuration (lane, hard shoulder and central reservation widths; safety barrier loca-
tion with regard to hard shoulders, etc.);
4. study of the highway flow conditions, carried out on the basis of the results of the analyses detailed in Section 4 and, in
addition, obtained in previous researches (Mauro, 2003, 2005, 2007).

We then proceeded with a quantitative estimation of the effects deriving by the HSR activation with the help of the pre-
dictive models in the Highway Safety Manual HSM (2010), which represents a useful support tool for decision making. In
accordance with the procedures in the Manual we determined for each of the 25 segments (see Table 12):

1. the safety performance functions (SPF): functions to estimate the expected average crash frequencies in base conditions
with regard to the annual average daily traffic on the highway segments in question;
2. the cash modification factors (CMF): frequency correction factors calculated with SPF (in base conditions) to take the real
geometric Highway A22 configuration into account (lane and hard shoulder widths, barrier presence, curve segment val-
ues, safety barrier location, etc.);
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 175

Table 12
A22 basic segments analysed.

N km Initial km Final Length [km] Name of the basic segment AADT [veh/day]
1 0+000 15+870 15.87 Brennero - Vipiteno 41.907
2 15+870 38+030 22.16 Vipiteno - Bressanone 44.159
3 38+030 47+600 9.57 Bressanone - Bressanone Z.I. 48.388
4 47+600 53+070 5.47 Bressanone Z.I. - Chiusa 46.896
5 53+070 77+470 24.4 Chiusa - Bolzano nord 50.362
6 77+470 85+330 7.86 Bolzano nord - Bolzano sud 48.778
7 85+330 101+800 16.47 Bolzano sud - Egna Ora 57.553
8 101+800 121+450 19.65 Egna Ora - S.Michele Mezz. 58.810
9 121+450 131+440 9.99 S. Michele Mezz. - Trento nord 57.865
10 131+440 136+460 5.02 Trento nord - Trento centro 52.464
11 136+460 142+000 5.54 Trento centro - Trento sud 52.874
12 142+000 157+850 15.85 Trento sud - Rovereto nord 61.066
13 157+850 166+740 8.89 Rovereto nord - Rovereto sud 60.366
14 166+740 179+125 12.385 Rovereto sud - Ala Avio 62.075
15 179+125 206+670 27.545 Ala Avio - Affi 62.646
16 206+670 225+370 18.7 Affi - Verona nord 48.216
17 225+370 228+000 2.63 Verona nord - int. aut. A4 60.431
18 228+000 243+670 15.67 Int. aut. A4 - Nogarole Rocca 62.675
19 243+670 256+180 12.51 Nogarole Rocca - Mantova nord 61.054
20 256+180 265+000 8.82 Mantova nord - Mantova sud 61.197
21 265+000 276+710 11.71 Mantova sud - Pegognaga 61.669
22 276+710 285+630 8.92 Pegognaga - Reggiolo Rolo 54.637
23 285+630 302+175 16.545 Reggiolo Rolo - Carpi 55.440
24 302+175 312+150 9.975 Carpi - Campogalliano 59.668
25 312+150 313+085 0.935 Campogalliano - raccordo Autosole 60.706

3. the expected average crash frequencies per unit for each route segment, precisely:
 accident frequency per year and per kilometre with a single vehicle involved and damage only to property;
 accident frequency per year and per kilometre with a single vehicle involved and seriously or fatally injured people;
 accident frequency per year and per kilometre with multiple vehicles involved and damage only to property;
 accident frequency per year and per kilometre with multiple vehicles involved and seriously or fatally injured people;
4. frequencies per segment (unit values per segment length) and the total A22, in case of HSR non-active;
5. frequencies per basic segment (unit values per segment length, cfr. Table 12) and the total A22, in case of HSR active;
6. calculation of absolute and percentage variations in crash frequencies concerning the previous two cases (with or without
the HSR).

The predictive model for these route segments is based on the following analytical expressions:

Np;fs;n;at;as Np;fs;n;mv;fi Np;fs;n;sv;fi Np;fs;n;mv;pdo Np;fs;n;sv;pdo 17

Np;fs;n;mv;fi C fs;ac;mv;fi  Nspf;fs;n;mv;fi  CMF1;fs;ac;mv;fi  . . . :CMFm;fs;ac;mv;fi  CMF1;fs;ac;at;fi  . . . :CMFm;fs;ac;at;fi 18

Np;fs;n;sv;fi Cfs;ac;sv;fi  Nspf;fs;n;sv;fi  CMF1;fs;ac;sv;fi  . . . :CMFm;fs;ac;sv;fi  CMF1;fs;ac;at;fi  . . . :CMFm;fs;ac;at;fi 19

Np;fs;n;mv;pdo Cfs;ac;mv;pdo  Nspf;fs;n;mv;pdo  CMF1;fs;ac;mv;pdo  . . . :CMFm;fs;ac;mv;pdo  CMF1;fs;ac;at;pdo  . . . :CMFm;fs;ac;at;pdo


20

Np;fs;n;sv;pdo Cfs;ac;sv;pdo  Nspf;fs;n;sv;pdo  CMF1;fs;ac;sv;pdo  . . . :CMFm;fs;ac;sv;pdo  CMF1;fs;ac;at;pdo  . . . :CMFm;fs;ac;at;pdo


21

where:
Np,fs,n,y,z: predicted average crash frequency of a freeway segment with n lanes, crash type y (y = sv: single vehicle, mv:
multiple vehicle, at: all types), and severity z (z = fi: fatal and injury, pdo: property damage only, as: all severities)
(crashes/yr);
Nspf,fs,n,y,z: predicted average crash frequency of a freeway segment with base conditions, n lanes, crash type y (y = sv:
single vehicle, mv: multiple vehicle, at: all types), and severity z (z = fi: fatal and injury, pdo: property damage only)
(crashes/yr);
CMFm,fs,ac,y,z: crash modification factor for a freeway segment with any cross section ac, feature m, crash type y (y = sv:
single vehicle, mv: multiple vehicle, at: all types), and severity z (z = fi: fatal and injury, pdo: property damage only);
Cfs,ac,y,z: calibration factor for freeway segments with any cross section ac, crash type y (y = sv: single vehicle, mv: multiple
vehicle, at: all types), and severity z (z = fi: fatal and injury, pdo: property damage only).
176 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Equation (17) shows that the expected accident frequency is given by the sum of four distinct rates: (a) accidents with
multiple vehicles involved and fatal and non-fatal casualties; (b) accidents with a single vehicle involved and fatal and
non-fatal casualties; (c) accidents with multiple vehicles involved and damage only to property; (d) accidents with a single
vehicle involved and damage only to property. These frequency rates are to be calculated with relations (18)(21). Once we
obtained the expected average crash frequencies per kilometre in base conditions and calculated the values of crash modi-
fication factors concerning the real geometric configuration of each motorway segment, we determined the expected average
crash frequencies per kilometre per segment at varying AADT.
On the basis of the planimetric layout (straight road length, circular curve radius and length, etc.), have been determined
the crash modification factors for the presence of planimetric curves (CMF1):
"  2 #
X
m
5:730
CMF1;fsac y z 1 a   Pc;i 22
i1
Ri

In which:

 CMF1, fs, ac, y, z: Crash modification factor for horizontal curvature in a freeway segment with any cross section ac;
 m = number of horizontal curves in the segment;
 Ri = equivalent radius of curve i (=[0.5/R2a,i + 0.5/R2b,i]0.5 if both roadbeds are curved, Ra,i if only one roadbed is curved) (ft);
 Ra,i = radius of curve i in one roadbed (ft);
 Rb,i = radius of curve i in second roadbed (used if both roadbeds are curved) (ft); and
 Pc, i = proportion of effective segment length with curve i.

CMF1 values for each base route segment are illustrated in Table 13. Table 14 reports CMF2CMF11 values.
By way of an example we mention the - Expected average crash frequency-AADT curves for the segment N.10 (Trento
North Trento Centre) distinguished for non-active HSR (4 lanes in total, see Fig. 18) and for active HSR (6 lanes in total,
see Fig. 19).
As to these curves, the frequency values to take into account are the AADT values recorded on the examined segment in
2015 (e.g. for segment n. 10 Trento North - Trento South, the AADT is equal to 52,464 veh/day, see Table 12).
The HSM recommends deriving calibration factors using randomly selected 3050 roadway sites that experienced a min-
imum of 100 crashes per year (Alluri et al., 2016). Currently no calibration factors have been studied in Italian highways.
Thus, in order to calibrate the model we processed crash data recorded on A22 highway in 2013, 2014 and 2015 (Fig. 20).
On the whole, there were:

 877 accidents in 2013;


 958 accidents in 2014;
 883 accidents in 2015.

Table 13
Crash modificator factors for horizontal curve.

N Segment Length (km) CMF1,fs,ac,mv,fi CMF1,fs,ac,mv,pdo CMF1,fs,ac,sv,fi CMF1,fs,ac,sv,pdo


1 Brennero - Vipiteno 15.87 1.08595 1.16991 1.35931 1.31283
2 Vipiteno - Bressanone 22.16 1.06988 1.13813 1.29211 1.25433
3 Bressanone - Bressanone Z.I. 9.57 1.05354 1.10584 1.22382 1.19487
4 Bressanone Z.I. - Chiusa 5.47 1.05818 1.11502 1.24322 1.21176
5 Chiusa - Bolzano nord 24.4 1.13060 1.25816 1.54594 1.47532
6 Bolzano nord - Bolzano sud 7.86 1.08508 1.16819 1.35567 1.30967
7 Bolzano sud - Egna Ora 16.47 1.00893 1.01766 1.03734 1.03251
8 Egna Ora - S.Michele Mezz. 19.65 1.00929 1.01836 1.03882 1.03380
9 S. Michele Mezz. - Trento nord 9.99 1.01026 1.02028 1.04289 1.03734
10 Trento nord - Trento centro 5.02 1.01298 1.02565 1.05425 1.04723
11 Trento centro - Trento sud 5.54 1.01530 1.03025 1.06396 1.05569
12 Trento sud - Rovereto nord 15.85 1.01489 1.02944 1.06226 1.05421
13 Rovereto nord - Rovereto sud 8.89 1.02255 1.04458 1.09428 1.08209
14 Rovereto sud - Ala Avio 12.39 1.02931 1.05794 1.12252 1.10668
15 Ala Avio - Affi 27.55 1.01473 1.02911 1.06156 1.05360
16 Affi - Verona nord 18.7 1.01064 1.02103 1.04446 1.03871
17 Verona nord - int. aut. A4 2.63 1.00581 1.01148 1.02427 1.02113
18 Int. aut. A4 - Nogarole Rocca 15.67 1.00368 1.00727 1.01537 1.01338
19 Nogarole Rocca - Mantova nord 12.51 1.00252 1.00497 1.01052 1.00916
20 Mantova nord - Mantova sud 8.82 1.00146 1.00288 1.00609 1.00531
21 Mantova sud - Pegognaga 11.71 1.00145 1.00287 1.00608 1.00529
22 Pegognaga - Reggiolo Rolo 8.92 1.00445 1.00879 1.01859 1.01619
23 Reggiolo Rolo - Carpi 16.55 1.00140 1.00276 1.00584 1.00508
24 Carpi - Campogalliano 9.975 1.00246 1.00487 1.01030 1.00897
25 Campogalliano - raccordo Autosole 0.935 1.00325 1.00642 1.01357 1.01181
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 177

Table 14
CMFs values.

CMF description CMF variable 4 lanes HSR (6 Lanes)


Lane width CMF2 0.989 0.989
Inside shoulder width CMF3, fs, ac, mv, fi 1.93 1.093
CMF3, fs, ac, mv, pdo 1.082 1.082
CMF3, fs, ac, sv, fi 1.093 1.093
CMF3, fs, ac, sv, pdo 1.082 1.082
Median width CMF4, fs, ac, mv, fi 1.150 1.150
CMF4, fs, ac, mv, pdo 1.144 1.144
CMF4, fs, ac, sv, fi 1.048 1.048
CMF4, fs, ac, sv, pdo 1.143 1.143
Median barrier CMF5, fs, ac, mv, fi 1.173 1.173
CMF5, fs, ac, mv, pdo 1.229 1.229
CMF5, fs, ac, sv, fi 1.173 1.173
CMF5, fs, ac, sv, pdo 1.229 1.229
High volume CMF6, fs, ac, mv, fi 1.016 1.002
CMF6, fs, ac, mv, pdo 1.013 1.002
CMF6, fs, ac, sv, fi 0.997 1.000
CMF6, fs, ac, sv, pdo 0.972 0.997
Lane change CMF7, fs, ac, mv, z 1.000 1.000
Outside shoulder width CMF8, fs, ac, sv, fi 1.089 1.000
CMF8, fs, ac, sv, pdo 1.089 1.000
Shoulder rumble strip CMF9, fs, ac, sv, fi 1.000 1.000
Outside clearance CMF10, fs, ac, sv, fi 1.040 1.090
Outside barrier CMF11, fs, ac, sv, fi 1.012 1.142
CMF11, fs, ac, sv, pdo 1.015 1.187

Fig. 18. Expected average crash frequency. Segment Trento Nord Trento Centro 4 lanes (without HSR).

The crashes were classified as required by the HSM model, thus determining:

 the yearly crash frequency per kilometre with a single vehicle involved and damage only to property;
 the yearly crash frequency per kilometre with a single vehicle involved and seriously or fatally injured people;
 the yearly crash frequency per kilometre with multiple vehicles involved and damage only to property;
 the yearly crash frequency per kilometre with multiple vehicles involved and seriously or fatally injured people.
178 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Fig. 19. Expected average crash frequency. Segment Trento Nord Trento Centro 6 lanes (with HSR).

Fig. 20. observed crash frequency for each base segment (year 2015).

We first determined the absolute frequencies of crashes occurred in the 25 base segments (cfr. Table 12) into which the
Brenner Highway was subdivided (see Fig. 20).
The calibration factor for each segment was determined using the following expression:
Pnsites Pnc
j1 No;wi;xi;y;z;j
Cw;x;y;z Pni1 Pn c 23
j1 Np;wi;xi;y;z;j
sites
i1

In which:

Cw, x y, z = calibration factor to adjust SPF for local conditions for site type w, cross section or control type x, crash type y,
and severity z;
nsites = number of sites;
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 179

nc = number of years in the crash period (yr);


No, w(i), x(i), y, z, j = observed crash frequency for site i with site type w(i) and year j (includes cross section or control type x
(i) for crash type y and severity z) (crashes/yr);
Np, w(i), x(i), y, z, j = predicted average crash frequency for site i with site type w(i) and year j (includes cross section or con-
trol type x(i) for crash type y and severity z) (crashes/yr).

The calibrated model made it possible to estimate, in the hypothetical case of continuous activation of the third dynamic
lane HSR (that is, for all hours in a year), the following average frequency variations in crashes on the A22 (weighted aver-
age; weight: the segment length), with regard to the current regulation of vehicle flows (non-active dynamic lane):

 percentage variation in yearly frequency per kilometre with a single vehicle involved and damage only to property:
D = +6.8%;
 percentage variation in yearly frequency per kilometre with a single vehicle involved and fatal and non-fatal casualties:
D = +12.5%;
 percentage variation in yearly frequency per kilometre with multiple vehicles involved and damage only to property:
D = 10.9%;
 percentage variation in yearly frequency per kilometre with multiple vehicles involved and fatal and non-fatal casualties:
D = 5.6%;
 percentage frequency variation in all types of crashes: D = 1.49%.

The fact that some crash types increase and others decrease is consistent with what was empirically observed in existing HSR
highways (Aron et al., 2010).
In the real working conditions, the HSR system is supposed to be activated exclusively during peak traffic hours. There-
fore, the estimation of the expected crash frequencies and percentage variations between the two analysed configurations
with or without active dynamic lane (HSR) needs to be based on hypotheses concerning the hours in a year with such
extremely high traffic peaks to cause congestion phenomena on Highway A22, thus requiring the HSR system activation.
Moreover, these traffic peaks could not simultaneously affect the entire A22 motorway along which the dynamical lane
should be created (as previously underlined, in the study hypotheses the HSR is 128 km long).
In the light of the previous observations, we examined the real peak traffic conditions recorded on Highway A22 (pro-
cessed data) or, in other words, the number of annual hours in which the flows reach or overcome the established alerting
values. The latter can be matched to two specific situations:

 flows corresponding (or near) to capacity values of lanes or carriageways (see Figs. 4 and 5 and Tables 58);
 flows determining the attainment of prefixed values of Levels of Service (LOS) which are thought by the management
body as minimum requirements.

Since it was shown (Mauro, 2005) that on the A22 only in 30 h a year the carriageway flows overcome 3000 pcu/h, has
been carried out a specific sensitivity analysis of the expected crash frequencies at varying number of hours which induce to
activate the highway traffic management device. We considered the following threshold levels:

 activation for 30 h a year (equal to 0.34% of the annual hours), which correspond to carriageway flows Q P 3.000 pcu/h;
 activation for 200 h a year (equal to 2.3% of the annual hours), which correspond to carriageway flows Q P 2.400 pcu/h.

Table 15 illustrates the expected crash frequencies in case of continuous activation of the third dynamic lane HSR (acti-
vation segment from km 102+00 to km 230+00).
Fig. 21 shows the percentage variations in expected crash frequencies regarding the two examined cases (active and non-
active HSR) at varying number of annual HSR operating hours. The percentage variations in crash frequencies linearly
increase with the annual hour number during which HSR is supposed to be activated. With the activation of 200 h per year,
the variation in crash frequency on the Highway A22 (weighted average) reaches at most 0.28% value (increase in crashes
involving single vehicles, with injuries and/or fatalities of people). They are negligible variations, of little significance. In
other words, the HSM model shows that the activation of the HSR system is not supposed to involve significant variations
in general safety conditions of the Highway A22.
The HSR system is activated with high values of the traffic demand; if coupled with a VSL (variable speed limit) system,
the flow regime tends to be homotachic; such conditions make it possible for vehicle flows to reach the maximum stability
levels (Mauro, 2005). Moreover, their corresponding capacity and density volumes, unlike those obtained by the relation
v = v(k) (see Fig. 3), are affected by very limited (at worst zero) dispersions and we can generally expect higher values of
critical density kjam than those provided by the fundamental diagrams (see Figs. 4, 5 and 16). In the absence of direct deter-
minations of capacity C, density kjam and velocity vjam in controlled flow regime (in respect of speed limits), we assume
kjam = 25 pcu/km/lane for all the lanes and kjam = 75 pcu/km/carriageway for the carriageway. The capacity values at varying
speed limit are given in Table 16.
The laws v = v(k) = v = cost; q = v  k; k = q/v can be easily deduced by examining the values in Table 16.
180 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

Table 15
Predicted average crash frequency. Continuous activation of the HSR.

N km Initial km Final Length [km] mv, fi mv,pdo sv, fi sv, pdo All types
Np,fs,46,mv,fi Np,fs,46,mv,pdo Np,fs,46,sv,fi Np,fs,46,sv,pdo Np,fs,46,at,as
Predicted average crash frequency [crashes/year]
1 0+000 15+870 15.87 3.0 25.3 3.7 11.0 43.0
2 15+870 38+030 22.16 8.3 20.7 4.7 16.7 50.3
3 38+030 47+600 9.57 2.3 7.0 1.0 4.7 15.0
4 47+600 53+070 5.47 2.3 5.7 2.0 4.7 14.7
5 53+070 77+470 24.4 16.7 34.0 5.0 25.3 81.0
6 77+470 85+330 7.86 4.3 16.3 2.0 14.0 36.7
7 85+330 101+800 16.47 8.3 13.3 1.7 16.1 39.3
8 101+800 121+450 19.65 10.3 10.6 5.7 19.1 45.7
9 121+450 131+440 9.99 4.0 11.3 1.3 13.7 30.3
10 131+440 136+460 5.02 2.7 4.3 0.7 3.7 11.3
11 136+460 142+000 5.54 2.3 5.6 1.0 4.0 13.0
12 142+000 157+850 15.85 8.3 9.9 3.4 9.4 31.0
13 157+850 166+740 8.89 4.7 10.9 1.3 8.0 25.0
14 166+740 179+125 12.385 6.0 7.0 3.0 9.0 25.0
15 179+125 206+670 27.545 11.3 25.9 5.4 24.1 66.6
16 206+670 225+370 18.7 8.3 25.5 4.7 15.7 54.3
17 225+370 228+000 2.63 2.3 5.3 1.0 8.0 16.7
18 228+000 243+670 15.67 11.3 15.3 7.0 18.7 52.3
19 243+670 256+180 12.51 5.7 9.7 7.3 19.3 42.0
20 256+180 265+000 8.82 4.7 5.0 2.0 11.0 22.7
21 265+000 276+710 11.71 6.7 13.3 7.3 25.3 52.7
22 276+710 285+630 8.92 4.3 6.0 6.7 13.3 30.3
23 285+630 302+175 16.545 6.3 12.7 8.3 24.0 51.3
24 302+175 312+150 9.975 13.3 15.0 4.0 17.3 49.7
25 312+150 313+085 0.935 0.0 2.3 0.3 3.3 6.0

Fig. 21. Percentage variations in expected crash frequencies regarding the two examined cases (active and non-active HSR).

Table 16
Capacity values as function of the speed limits v.

v [km/h] kjam [pcu/km/lane] C [pcu/h]


Lane (central lane, passing lane, right lane) 60 25 1500
70 25 1750
80 25 2000
Carriageway 60 75 4500
70 75 5250
80 75 6000

Therefore, it goes without saying that the adoption of appropriate speed limits can increase the infrastructure capacity
and/or reduce the number of lane changes made by the drivers. Moreover we should consider that:

(a) if the velocity processes identified in series of instantaneous speeds (see Figs. 1214) present absolute values M = |b| in
the angle coefficient of the relation r2 = a + bln(k) (see Table 10) higher than 4 [m2 km s2], the speed limit restriction
between 80 km/h and 100 km/h - when the reliability / (see Eq.(8)) reaches the critical interval (/ = 0.900.95) leads
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 181

Fig. 22. Example of coordinated set of control programs for the motorway segments.

to a significant reduction in this coefficient, thus indicating a greater regularity in random processes of the instanta-
neous speeds (Mauro, 2005; Mauro et al., 2013). It follows an increase of the carriageway capacity or, traffic demand
being equal, a marked increase in the system reliability. For instance, in the German highway A14, as a consequence of
the said speed limits, the capacity values were shifted from the order of 33004000 veh/h, with a reduction in the
average M value from 7 [m2 km s2] to 3 [m2 km s2] (Cremer and Fleischmann, 1987);
(b) if M is determined with values lower than 4 [m2 km s2], the speed limit restriction does not increase either the road
capacity or the traffic flow reliability /, but after the imposition there is a reduction in lane changes. Such a circum-
stance contributes to avoid an increase in M and, consequently, a reduction in / value. For instance, such effects were
seen in experimental observations made on Dutch highways (Klijnhout, 1984). In the presence of M determinations,
on average, equal to M = 3.5 [m2 km s2], concomitant with high vehicle capacity values, the imposition of a speed
limit of the order of the average flow speed did not lead to the speed reduction, but the lane change resulted to be
more than halved (compared to the previous situation with no speed limit restrictions), with consequent stabilization
of the flow conditions.

Fig. 22 shows a possible coordinated set of control programs for the motorway segments (subsegments) with a length
ranging between 0.5 km and 1 km. On the basis of the determinations of the reliability / and speed process level V at the
centre Xi of the generic subsegment i, and of the intervals of these parameters (/, V) written in the upper line in
Fig. 22, the programme Pk, k = 0, 1, . . ., 4, is selected to determine the messages represented by the speed limits in this case
to be shown on the signal portal located in Xi and the portals in the adjacent segments.

6. Conclusions

The systems for the traffic flow management and control are frequently used in existing operational infrastructures to
improve their performance. The best-known and widely implemented nowadays are the ramp metering (RM), high occu-
pancy vehicle (HOV) lanes, real-time variable speed limits (VSLs), reversible lanes (RL) and hard-shoulder running (HSR).
This research examines the hard-shoulder running (HSR). Numerous studies underline that HSR produces significant
increases in capacity and reduction in travel times. Such devices are generally combined with VSLs. We examined the case
study of the HSR project on the motorway A22 (Brenner Motorway) 128-km long, and especially the variations in its safety
levels compared to the current configuration.
At first we had to assess the conditions of the existing operational infrastructure by determining the flow curves, capacity,
critical density, speed levels, reliability /, etc., in both the initial configuration (non-active HSR) and the project configuration
(active HSR). Considerable capacity increases have been estimated (e.g. for section San Michele the capacity moves from
3361 pcu/h to 4535 pcu/h a 35% increase for northbound roadway, and from 3340 pcu/h to 4246 pcu/h a 27% increase
for southbound roadway).
In addition, has been carried out a safety analysis of the newly-designed device by estimating the expected crash frequen-
cies with the HSM methodology (HSM, 2010).
The A22 was subdivided into 25 elementary segments, each with constant average daily traffic values (AADT). For each
segment have been obtained: (a) safety performance functions (SPF): (b) crash modification factors (CMF); (c) expected aver-
age unit frequencies per segment. For every segment the calibration factor was obtained by considering crash data occurred
in A22 in the years 20132015 (recorded data: 877 accidents in 2013; 958 accidents in 2014; 883 accidents in 2015). The
estimation of the expected crash frequencies also depends on the conditions of the peak traffic hours recorded in A22
182 M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183

(observed data), or on the annual hours when the flows reach or overcome present alarming values that require the HSR acti-
vation. Thus, has been done a sensitivity analysis of the expected frequencies of the accidents at varying hour number during
which the motorway traffic management device (HSR) is supposed to be used. Prudently, the following threshold values
were considered: (1) HSR activation for 30 h per year (equal to 0.34% of the annual hours) which correspond to carriageway
flows Q P 3000 pcu/h; (2) HSR activation for 200 h per year (equal to 2.3% of the annual hours) which correspond to car-
riageway flows Q P 2400 pcu/h. We estimated that the HSR activation for 200 h a year leads to negligible variations in
the expected crash frequencies, compared to the current time (at most 0.28% for single-vehicle accidents with fatal or
non-fatal casualties). Therefore, the application of the HSM model to the case study shows that the activation of the HSR
system, despite its considerable benefits on highway functionality, does not lead to significant variations in the general
safety conditions.
The management of speed limits (VSLs) on HSR must be based on the analyses of the processes of instantaneous speeds.
More precisely, if the values M = |b| of the angle coefficient of relation r2 = a + bln(k) are over 4 [m2 km s2], a speed limit
restriction between 80 km/h and 100 km/h, when the reliability / reaches the critical interval (/ = 0.900.95), determines
increases in carriageway capacity. On the contrary, if M assumes values below 4 [m2 km s2], the imposition of speed limits
does not increase either the road capacity or the traffic flow reliability /, but there is a reduction in lane changes as a con-
sequence of the speed limit restriction. In conclusion, on the basis of the previous observations, we have suggested a possible
set of coordinated speed control programs, based on HSR, for motorway segments with a length between 0.5 km and 1 km,
designed on the determination of reliability / and speed process level V:

Acknowledgments

The Authors wish to thanks Dr. Eng. Walter Pardatscher, CEO of the Autostrada del Brennero SpA, for the constant will-
ingness during the development of this research.

References

Alluri, P., Saha, D., Gan, A., 2016. Minimum sample sizes for estimating reliable Highway Safety Manual (HSM) calibration factors. J.Transp. Saf. Secur. 8 (1),
5674.
Aron, m., Cohen, S., Seidowsky, R., 2010. Two french hard-shoulder running operations: some comments on effectiveness and safety. In: 13th International
IEEE Annual Conference on Intelligent Transportation Systems Madeira Island, Portugal, September 1922.
Barcel, J., 2010. Fundamentals of Traffic Simulation. Springer.
BBC News World Edition. Creative ways to beat congestion. November 2004. <http://news.bbc.co.uk/2/hi/uk_news/magazine/4044803.stm>.
Brinckerhoff, P., 2010. Synthesis of Active Traffic Management Experiences in Europe and the United States. FHWA-HOP-10-031.
Carbaugh, J., Godbole, D.N., Sengupta, R., 1998. Safety and capacity analysis of automated and manual highway systems. Transp. Res. Part C: Emerg. Technol.
6C (12), 6999.
Cremer, M., Fleischmann, S., 1987. Traffic responsive control of freeway networks by a state feedback approach. Proceedings of the 10th Int. Symposium on
Transportation and Traffic Theory. Elsevier, new York.
FHWA, 2010. Efficient Use of Highway Capacity Summary. Report to Congress, 2010 Report No. FHWA-HOP-10-023.
Elefteriadou, L., 2014. An Introduction to Traffic Flow. Springer.
Ferrari, P., 1988. The reliability of motorway transport system. Transp. Res. Part B 22 (4), 291310.
Ferrari, P., 1991. The control of motorway reliability. Transp. Res. Part A 25 (6), 419427.
FHW, 2006. Ramp Management and Control, Handbook.
Gerlough, D.L., Huber, M.J., 1975. Traffic flow theory: a monograph. TRB special report 165.
Gross, F., Eccles, K., Jovanis, P.P., Chen, K., 2009. Safety Evaluation of Lane and Shoulder Width Combinations on Rural, Two-Lane, Undivided Roads, FHWA
Tech Brief, Publication No. FHWA-HRT-09-032, Federal Highway Administration.
Hamouda, O., Kaniche, M., Kanoun, K., 2009. Safety modeling and evaluation of automated highway systems. In: Proceedings of the International
Conference on Dependable Systems and Networks.
Highway Capacity Manual (HCM 2010), 2010.
Highway Safety Manual (HSM), American Association of State Highway Transportation Officials (AASHTO), 2010.
Ioannou, P.A. et al, 1997. Automated Highway Systems. Springer Science+Business media, LLC.
Klijnhout, J.J., 1984. Motorway control and signalling: the test of the time. Traffic Eng. Control 25.
Kuhn, B., 2010. Efficient use of highway capacity summary. Report No. FHWA-HOP-10-023. Texas Transportation Institute, Texas A&M University, College
Station, TX.
Khne, R., Michalopoulos, P., Zhang, HM., 1992. Continuum flow models, Chapter 5 in Traffic flow theory, <http://www.tfhrc.gov>.
Lee, C., Hellinga, B., Saccomanno, F., 2006. Evaluation of variable speed limits to improve traffic safety. Transp. Res. Part C: Emerg. Technol. 14 (3), 213228.
Lemke, K., Irzik, M., 2006. Temporary Use of Hard Shoulders-Experiences and Economic Assessment. Federal Highway Research Institute, Traffic Planning,
Highway Design, and Safety Analyses Section, Bergisch Gladbach, Germany.
Lemke, K., Moritz, K., 2001. Freigabe von Seitenstreifen an Bundesautobahnen. Berichte der Bundesanstalt fr Straenwesen, Verkehrstechnik Heft V94,
Bergisch Gladbach.
Li, Y., Chow, A.H.F., Cassel, D.L., 2014. Optimal control of motorways by ramp metering, variable speed limits, and hard-shoulder running. Transp. Res. Rec.
2470, 122130.
Michael, J.B., Godbole, D.N., Lygeros, J., Sengupta, R., 1998. Capacity analysis of traffic flow over a single-lane automated highway system. ITS J. 4, 4980.
Mauro, R., 2003. Analisi di traffico, elaborazione di modelli e sistemi per la stima dellaffidabilit per lAutostrada A22, Italia [Traffic analysis, development of
models and systems for estimating reliability on the A22 Freeway, Italy], Technical report, Part I, Autostrada del Brennero, Trento, Italy (in Italian).
Mauro, R., 2005. Analisi di traffico, elaborazione di modelli e sistemi per la stima dellaffidabilit per lAutostrada A22, Italia [Traffic analysis, development of
models and systems for estimating reliability on the A22 Freeway, Italy], Technical report, Part II, Autostrada del Brennero, Trento, Italy (in Italian).
Mauro, R., 2007. Analisi di traffico, elaborazione di modelli e sistemi per la stima dellaffidabilit per lAutostrada A22, Italia [Traffic analysis, development of
models and systems for estimating reliability on the A22 Freeway, Italy], Technical report, Part III, Autostrada del Brennero, Trento, Italy (in Italian).
Mauro, R., Giuffr, O., Gran, A., 2013. Speed stochastic processes and freeway reliability estimation: evidence from the A22 freeway, Italy. J. Transp. Eng.
139 (12), 12441256.
Mauro, R., 2015. Traffic and Random Processes. An Introduction. Springer.
M. Guerrieri, R. Mauro / Transportation Research Part A 92 (2016) 162183 183

Mattheis, C., 2002. (Ingenieurbro Dipl. Ing. H. Vssing GmbH, Dsseldorf), Auswirkungen der Umnutzung von BAB-Standstreifen, Berichte der
Bundesanstalt fr Straenwesen, Verkehrstechnik Heft V91, Bergisch Gladbach, Januar 2002.
May, A.D., Keller, H.E.M., 1967. Non-integer car-Following Models. Highw. Res. Rec. 199, 1932.
May, A.D., 1990. Traffic Flow Fundamentals. Prentice-Hall.
Moritz, K., Wirtz, H., 2003. Auswirkungen von Standstreifen-numnutzungen auf dem Straenbetriebdienst. Berichte der Bundesanstalt fr Straenwesen,
Verkehrstechnik Heft V107, Bergisch Gladbach, September 2003.
NCHRP SYNTHESIS 432, 2012. Recent Roadway Geometric Design Research for Improved Safety and Operations. Transportation Research Board,
Washington, D.C.
Papacostas, C.S., Prevedouro, P.D., 2001. Transportation Engineering and Planning, third ed. Prentice Hall.
Papageorgiou, M., Hadj-Salem, H., Blosseville, J.-M., 1991. ALINEA: a local feedback control law for on-ramp metering. In: Transportation Research Record
1320. TRB, National Research Council, Washington, D.C., pp. 5864.
Taale, H., 2006. Regional Traffic Management Method and Tool. AVV Transport Research Centre, Rotterdam, The Netherlands: Ministry of Transport, Public
Works, and Water Management, Directorate-General of Public Works and Water Management. Presentation to Planning for Congestion Management
Scan Team.
Wang, C., David, B., Chalon, R., Yin, C., 2016. Dynamic road lane management study. A Smart City application. Transp. Res. Part E: Log. Transp. Rev. 89,
272287.
Wang, H., Li, J., Chen, Q.Y., Ni, D., 2009. Speed-density relationship: from deterministic to stochastic. In: TRB 88th Annual Meeting at Washington D. C.