Anda di halaman 1dari 20

Transportation Research Part C 56 (2015) 1833

Contents lists available at ScienceDirect

Transportation Research Part C

j o u r n a l h o m e p a g e : w w w . e l s ev i e r . c o m / l o c a t e / t r c

A continuous-flow-intersection-lite design and traffic control for


oversaturated bottleneck intersections
a a a,b a,b,
Weili Sun , Xinkai Wu , Yunpeng Wang , Guizhen Yu
a School of Transportation Science and Engineering, Beihang University, Beijing, China
bBeijing Key Laboratory for Cooperative Vehicle Infrastructure Systems and Safety Control, Beihang University, Beijing, China

article info abstract

Article history: Oversaturation has become a severe problem for urban intersections, especially the bottle-neck intersections
Received 10 November 2014 that cause queue spillover and network gridlock. Further improvement of oversaturated arterial traffic using
Received in revised form 7 February 2015 traditional mitigation strategies, which aim to improve intersection capacity by merely adjusting signal
Accepted 8 March 2015 Available online 29
control parameters, becomes challenging since exiting strategies may (or already) have reached their
March 2015
theoretical limits of optimum. Under such circumstance, several novel unconventional intersection
designs, including the well-recognized continuous flow intersection (CFI) design, are originated to improve
Keywords:
the capacity at bottleneck intersections. However, the requirement of installing extra sub-intersections in a
Oversaturation
CFI design would increase vehicular stops and, more critically, is unacceptable in tight urban areas with
Unconventional intersection design
Continuous flow intersection closed spaced intersections. To address these issues, this research proposes a simplified continuous flow
Capacity analysis intersection (called CFI-Lite) design that is ideal for arterials with short links. It benefits from the CFI
Signal control concept to enable simultaneous move of left-turn and through traffic at bottleneck intersections, but does not
Multi-objective mixed integer programming need installation of sub-intersections. Instead, the upstream intersection is utilized to allocate left-turn traffic
to the displaced left-turn lane. It is found that the CFI-Lite design performs superiorly to the conventional
design and regular CFI design in terms of bottleneck capacity. Pareto capacity improvement for every traffic
stream in an arterial system can be achieved under effortless conditions. Case study using data collected at
Foothill Blvd in Los Angeles, CA, shows that the new design is beneficial in more than 90% of the 408
studied cycles. The testing also shows that the average improvements of green bandwidths for the
synchronized phases are significant.

2015 Elsevier Ltd. All rights reserved.

1. Introduction

With traffic congestion continuing to grow in urban areas around the world, more and more signalized intersections are operated under
oversaturated conditions. Under the condition of oversaturation, typical traffic control strategies do not work as efficiently as necessary. As
indicated by the 2011 Traffic Signal Operation Self Assessment surveys (NTOC, 2011), the majority of agencies involved traffic signal operation
and maintenance are already stretched thin and challenged to provide adequate service to drivers in their jurisdictions. Oversaturated conditions
present an additional burden for practitioners that do not have adequate tools for addressing such situations (NCHRP, 2012).

Corresponding author at: School of Transportation Science and Engineering, Beihang University, Beijing, China. E-mail address:
yugz@buaa.edu.cn (G. Yu).

http://dx.doi.org/10.1016/j.trc.2015.03.011
0968-090X/ 2015 Elsevier Ltd. All rights reserved.
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 19

Numerous efforts have been made to address oversaturation. Over the years, researchers have realized that the onset of oversaturation is
usually caused by vehicle queues first occurring at some bottleneck intersections (i.e. critical intersections) and then propagating upstream and
sprawling all over the arterial network, causing so-called network gridlock (Gazis, 1964; Abu-Lebdeh and Benekohal, 2000; Daganzo, 2007; Wu
et al., 2010). Therefore, an efficient way to mitigate oversaturation is to improve the utilization of the capacity at bottleneck intersections in order
to prevent queue sprawling and hasten recovery after oversaturation. Aligning with this concept, some classical mitigation strategies, such as
throughput maximiza-tion, queue balancing, negative offset, and metering and gating, have been proposed for many years and have successfully
relieved arterial congestion at some level (e.g. Gazis, 1964; Michalopoulos and Stephanopoulos, 1977a,b; Chang and Lin, 2000; Park et al.,
2000; Abu-Lebdeh and Benekohal, 2000; Lieberman et al., 2000; Liu and Chang, 2011; Hu et al., 2013).

However, with traffic continuing to grow significantly, further improvement of arterial traffic using traditional mitigation strategies has
become challenging; simply because these strategies, which improve intersection capacity utilization by merely adjusting signal control
parameters, could (or already) have reached their theoretical limits. As signal control essentially is an optimization problem, there must be a
(or several) theoretical optimal solution(s), which, once reached, can-not be improved anymore. To address this issue, instead of purely adjusting
signal timing, more and more researchers begin to seek some novel intersection geometric designs, which can be used to further resolve the
problem of urban traffic congestion.
Several novel geometry designs (i.e. unconventional intersection designs) have been proposed recently to improve the capacity utilization at
bottleneck intersections. Most of these designs are originated to remove the conflict between left-turn and opposite through movements since
left-turn movement is one of the main causes which contribute to the inefficiency of conventional at-grade intersections. For example, some
designs, such as median U-turn, simply eliminate left turns by requiring drivers to first travel straight through the main intersection and then
execute their left turns by making U-turns at downstream (El Esawey and Sayed, 2013). This design reduces the total number of phases therefore
increasing the intersection capacity. The drawbacks, however, are that vehicles in some traffic stream are rerouted to experience long travel
distance, which not only wastes more time and fuel, but also increases the demand for the intersection (Autey et al., 2013). Other designs, such
as tandem intersection design, reorganize the left-turn and through traffic in tandem before the intersection, so that all travel lanes are utilized for
both left-turn and through traffic (Xuan et al., 2011). Nevertheless, lane blocking may happen if the anterior movement in tandem is somehow
unable to be fully discharged.

One of the well-recognized unconventional intersection designs is the continuous flow intersection (CFI), which moves the left-turn and
through movements simultaneously by adjusting the locations of left-turn or through lanes (Goldblatt et al., 1994). In a CFI design, the green
time initially allocated to the left-turn and through phases can be summed up together to serve all traffic therefore significantly improving
intersection capacity. The CFI design was reported to outperform other unconventional designs (Reid and Hummer, 2001; Jagannathan and
Bared, 2005; El Esawey and Sayed, 2007; Hughes et al., 2010; Autey et al., 2013; Wu et al., 2014), However, the CFI design requires installation
of new sub-intersections, which may generate more vehicular stops (Yang et al., 2013). It was also found that to some extent, the bottleneck for
left-turn movement shifts from the main intersection to sub-intersections (Carroll and Lahusen, 2013). Moreover, the construction of sub-
intersections is costly, not only in terms of money, but also the large space required (Hughes et al., 2010). Typically, sub-intersections in a CFI
design are suggested to be installed 300 400 feet upstream the intersection (Bared, 2009; Hummer and Reid, 2000), while the link length
between adjacent intersections along an arterial in an urban city are usually around 1000 feet. Considering the short link lengths between
adjacent intersections in tight urban areas, constructing a CFI becomes almost impossible.

To conquer the aforementioned problems, this research proposes a simplified continuous flow intersections design, called CFI-Lite, to
improve the capacity at oversaturated bottleneck intersections. The new design uses the existing upstream intersection, instead of newly
constructed sub-intersection, to allocate left-turn traffic to displaced left-turn (DLT) lane. Thus left-turn traffic turns left (to the DLT lane) in
advance at the upstream intersection. The CFI-Lite design benefits from the CFI concept to enable simultaneous move of left-turn and through
traffic at the bottleneck intersection, but does not need the installation of extra pre-signal and construction of sub-intersections. Therefore, CFI-
Lite is ideal for tight urban areas where the link lengths between adjacent intersections are too short to construct regular CFIs. It is obvious that
problems generated from sub-intersections do not exist anymore. In addition, this research further seeks for a Pareto capacity improvement for a
CFI-Lite design. A multi-objective mixed integer programming problem is formulated and, by solving this problem, the systematically
improvement of the capacity of bottleneck intersections can be achieved.

The rest of the paper is organized as follows. Section 2 describes the design concept of CFI-Lite design. The comparative analysis of capacity
for the conventional design, the regular CFI design and the CFI-Lite design is conducted in Section 3. Section 4 formulates the traffic control
problem at CFI-Lite intersections as a multi-objective mixed integer programming problem and seeks for Pareto improvement. Section 5 briefly
introduces how the CFI-Lite design is expanded to address consecutive bottlenecks. A case study using the data collected from Foothill Blvd in
the city of Los Angeles, CA is shown in Section 6. Section 7 concludes the paper.

2. Design concept

Fig. 1 presents the design concept of the CFI-Lite intersections. The objective of the CFI-Lite design is to remove the con-flicts between
heavy left-turn movements and opposing through movements at the bottleneck intersection along an arterial.
20 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

Fig. 1. The CFI-Lite intersection.

This has been realized by adjusting the position of left-turn lanes on the arterial. For each arterial approach connecting to the bottleneck
intersection, a DLT lane is placed to the left of opposing traffic lanes. At the same time, left turn is expressly prohibited in the regular lanes
which now only serve through and right-turn traffic at the bottleneck intersection. Vehicles from the arterial make left turn at the bottleneck
intersection by two steps: first divert to the DLT lane at upstream intersection, and then turn left to the side street at the bottleneck intersection.
Notice here we assume the upstream/ downstream intersections have a separate left turn phases which allow left turn get into DLTs.

The bottleneck intersection in the CFI-Lite design is the same as the primary intersection in a regular two-legged CFI design. The signal
phase is able to serve both through and left-turn traffic simultaneously. By doing so, the protected left-turn phase is no longer needed, leading
significant improvement of signal control efficiency. The only drawback of this design is that side street right-turn is not permitted on red
anymore; instead, a sign of No Right Turns On Red Light will be needed to protect the right turn traffic from side streets to major arterials.

The difference between CFI-Lite and regular CFI designs comes from the way how left-turn traffic crosses the arterial to the DLT lanes. In a
regular CFI design, additional sub-intersections are installed several hundred feet upstream the primary intersection, where cross-arterial left-turn
traffic and opposing through traffic are controlled by a two-phase signal. Such construction is no longer required in a CFI-Lite design. In a CFI-
Lite design, left-turn vehicles cross the arterial at the inter-section upstream of the bottleneck, and then drive into the DLT lane. But note that
right-turn movements on the opposing approach will be prohibited at the upstream intersection during the period when left-turn traffic crosses
the arterial. In addi-tion, the CFI-Lite design might bring some inconvenience for bus passengers when the bus stop is located beside the DLT
lane. Passengers dropping off buses have to cross an additional vehicular lane (i.e. the DLT lane) to reach the curb. Construction of median
refuge and crosswalk will be needed at the bus stop.

It has to be pointed out that, similar to regular CFI design, the CFI-Lite design could result in driver confusion due to unconventional
manipulations. Divers are expected to anticipate the left turn at the bottleneck intersection, and make indi-rect left turn to the DLT lane in
advance. Otherwise, it could lead to wrong-way movements and generate safety issues. But this situation can be avoided by providing adequate
signage, pavement markings and some warning messages. Nevertheless, a comprehensive assessment of drivers discomfort/comfort about the
CFI-Lite design will be needed. Previous related stud-ies on the regular CFI design might provide valuable reference (e.g. Abramson et al., 1995;
Hughes et al., 2010).

3. Capacity analysis

The CFI-Lite is naturally superior to the conventional intersection and regular CFI designs in terms of bottleneck capacity. To demonstrate
this point, we compare the intersection capacity values for three designs (conventional, CFI, and CFI-Lite) in this section. Since both CFI and
CFI-Lite are designed to resolve the conflicts between left-turn and opposing through move-ments at the bottleneck intersection, we focus on
analyzing the capacity for these two critical movements. A simple arterial system with three consecutive intersections, denoted by i 1, i, and i +
1, is discussed (see Fig. 2). Intersection i is treated as the bottleneck intersection.

First, the locations of the conflict points of the two critical movements (i.e. left-turn and opposing through movements) among three designs
are depicted in Fig. 2. As shown in the figure, in the conventional design, two conflict points (marked by red circle with cross symbol inside) for
these two movements (marked by dashed arrows) are located at the bottleneck inter-section i. In the CFI design (for two-legged), conflict points
to sub-intersections where the left-turn vehicles cross the arterial following pre-signals to
are shifted from intersection i
the DLT lanes. Such shift is beneficial because the spare time on the arterial is used to solve the conflicts, while
some green time is freed at the bottleneck intersection to serve more traffic. In the CFI-Lite design, the conflicts
are added to the conflicts of left-turn and opposite through movements at adjacent upstream
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 21

N
i-1 i
g6 g6 g6 i+1

i
g5 i-1 g1
i-1
g5 i g1 g5 i+1 g1
i+1

i-1
g2 g2 i g2
i+1

i-1 i i+1

(a) Conventional design

N Sub-intersection

G5i

G 6i
i-1 DLT i+1
G6
sub
G6sub G6
i-1 i-1 i+1 i+1
G1 G1 G1
G5 G5
i-1 G5sub G2i G2sub i+1

G2 DLT G2
i-1 G 1i
i i+1
Sub-intersection
(b) two-legged CFI design
N
DLT G5i

G 6i
i+1
i-1 G6
G6
G5i+1 G1 i+1
G2i
i-1 i-1
G5 G1 G2i+1
G2 i-1 DLT
i
G1
i-1 i i+1

(c) CFI-Lite design

Left-turn movement Through movement DLT: Displaced left-turn lane Conflict point

Fig. 2. Conflict points for three intersection designs: conventional, two-legged CFI, and CFI-Lite.

intersections i 1 and i + 1. Since the conflict points between left-turn and opposite through movements at intersections i 1 and i + 1 already exist,
technically the new CFI-Lite design removes the conflict points between left-turn and opposite through movements at intersection i without
generating any new conflict points.
With the shifting of the conflict points from the bottleneck intersection to upstream non-bottleneck intersections, the CFI-Lite gains more
capacity for the bottleneck intersection therefore performs superiorly to the conventional and regular CFI designs. Here we present a comparison
based on a simple scenario to give a qualitative insight why CFI-Lite provides more (at least not less) capacity than the other two designs.
Detailed quantitative discussion of the capacity improvement will be presented in the next section.

Assume that the discharging flow rate per lane and number of lanes for each phase are the same for all three intersections, therefore the
capacity for each traffic stream can be simply represented by the minimum effective green time experienced at these intersections. We
denominate phase 2 the eastbound (EB) through movement and other phases following NEMA prin-ciple for each intersection. For the ease of
exposition, single-ring signal timing plan with four phases: 2&6, 1&5, 4&8, and 3&7 is assumed in this comparison. Following notations are
used (also see Fig. 2):
22 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

gkj: the effective green time for phase j at intersection k initially set in the conventional design;
Gkj: the effective green time for phase j at intersection k in the CFI or CFI-Lite design;
Gsubj: the effective green time for phase j at the sub-intersection in the CFI design.

For the purpose of demonstration, the EB traffic traversing from intersection i 1 to intersection i is analyzed here. Note the exactly same
situation happens to the westbound (WB) traffic. The capacity analyses for three designs are presented as follows (also shown in Fig. 3).

(1) Capacity for the conventional intersection design: With a conventional intersection design, the capacity for EB through and left-turn
movements at bottleneck intersection i can be simply represented by their effective green times, g i2;6 and gi1;5, respectively (see Fig.
3(a)). Note that gi2;6 and gi1;5 are conflicting in the conventional design. More importantly, as intersection i is treated as a bottleneck, less
traffic from side streets arrives at upstream intersection i 1. Then, the green time for EB through movement should be larger at
intersection i 1 than that at intersection i. Therefore, gi2;61 P gi2;6 holds, and it is likely gi2;61 P gi2;6 gi1;5 will also hold.

(2) Capacity for the regular CFI design: For an intersection with a regular CFI design, g i2;6 and gi1;5 are summed together at intersection i to
serve through and left-turn movements simultaneously. So the capacity for the through and
left-turn traffic at bottleneck intersection could be improved to as high as G i2;6 Gi1;5 gi2;6 gi1;5 (Fig. 3(b)). But the bottleneck has
been, to some extent, shifted to the sub-intersection. This is because left-turn traffic can cross the arterial to the DLT lane only when
through movements on the cross streets (i.e. north and south) are given green, and then pass the bottleneck intersection when through
movements on the same road are given green. Therefore, the effective green for left-turn traffic is constrained by the minimum of the
above two. In other words, the bottleneck
capacity for the critical left-turn movement is constrained by G sub5, which is equal to or less than the green time for the through
movement on the cross streets, i.e., G sub5 6 gi4;8. The only way to increase left-turn capacity without shrinking the green time for side
street traffic is to allocate some green time from main street traffic to side street at bottleneck intersection. Thus the left-turn capacity
could be increased at the cost of reducing the same amount of through capacity (Fig. 3(c)). Nevertheless, such adjustment is not expected
because the through traffic on the arterial is usually more essential.

(3) Capacity for the CFI-Lite design: For intersections with the CFI-Lite design, the conflict of left-turn and through
movements is also removed at bottleneck intersection i, leading the potential capacity improvement to
G2i;6 G1i;5 g2i;6 g1i ;5 (Fig. 3(d)). But different with the regular CFI design, the conflict point is shifted to upstream

intersection i 1, so the bottleneck capacity for left-turn movement is restricted by Gi 1



gi 1 . Similar trade-off
1;5 1;5
between the green times serving left-turn and opposing through movements can be made to increase the bottleneck capacity for left-turn
movement. The most delightful feature of the CFI-Lite design is that the trade-off is made at intersection i 1 which is less saturated than
intersection i. Therefore, compared to a regular CFI design, the capacity of left-turn traffic, i.e. G i1;51, could be further increased by at
most d2 gi2;61 gi2;6 gi1;5 without reducing the through capacity (Fig. 3(e)).

From the analysis above, the CFI-Lite design shows obvious superiority to the conventional design and the regular CFI design. Notice CFI-
Lite could further gain some extra green time from the reduction of lost time or inter-green time because

Fig. 3. Capacity comparisons between different designs.


W. Sun et al. / Transportation Research Part C 56 (2015) 1833 23

fewer phases will be required for a CFI-Lite design compared to the conventional design. Nevertheless, several remarks regarding the superiority
need to be pointed out here:

The cases shown in Fig. 3 imply an assumption that gi2;61 P gi2;6 gi1;5. If not, the corresponding capacity improvement in the CFI or CFI-
Lite design would be smaller, and the left-turn capacity cannot be increased without the decrease of through capacity in the CFI-Lite design.
Fortunately, the inequality holds very likely for bottleneck intersections in an arterial system where major and minor roads intersect the
arterial alternately.
The left-turn capacity in the CFI-Lite design may be further improved by increasing the value of gi1;51 via lead-lag optimiza-tion at
intersection i 1. For example, if we want to favor one direction, the phases can be redesigned to 2&5 and 1&6 so that the green time for left-
turn and through traffic to the same direction could be increased at the same time. This amelioration is common for arterials along which tidal
traffic phenomenon is evident.
The left-turn capacity in the CFI-Lite design can also be increased by changing a through lane to left-turn lane at intersection i 1, as will be
discussed later in Section 4.

4. Signal control for the CFI-Lite intersections: a Pareto improvement

Because of the geometric change of a CFI-Lite design, signal timing for a CFI-Lite intersection has to be adjusted accordingly. This section
discusses the signal control strategy for a CFI-Lite intersection. Instead of minimizing total intersection delay or maximizing throughput under
given demand, we aim to find a Pareto improvement which shows the superior performance of the CFI-Lite design in comparison with the
conventional intersection design. To a set of individuals, a Pareto improvement is defined as a change that is an improvement for at least one and
no worse for any other. Therefore, given any existing signal timing plan for conventional designed intersections, we aim to improve the
bottleneck capacity of crucial streams with an adjusted timing plan in the context of CFI-Lite intersections, while keeping the capacity for any
other stream unchanged or improved as well.

4.1. The multi-objective mixed integer programming formulation

To seek for the Pareto improvement in the CFI-Lite design, we formulate the problem as a multi-objective mixed integer programming (MO-
MIP) problem. The problem is expressed using a three-intersection arterial corridor with the bottleneck intersection in the middle as the study
case shown in Fig. 2.

Decision variables: Given the existing signal timing plan {g kj} and number of lanes {n kj} for phases of all the intersections in the conventional
design, k i 1, i; i 1, j = 1, 2, . . . , 8, the decision variables are the corresponding green time {G kj} and number of lanes {Nkj} in the CFI-Lite
design. {Nkj} is a set of positive integers and {Gkj} is a set of positive real numbers.
Objective: The objective is to maximize the bottleneck capacity for crucial movements, including the left-turn and through movements for both
EB and WB. Therefore, multiple objectives are involved, i.e.

max J f 1x BET ; f 2x BWT ; f 3x BEL; f 4x BWL 1

where x is the vector of decision variables, f l(x) is the lth objective function, B denotes the bottleneck capacity in the CFI-Lite design.
Specifically, BET BWT BEL and BWL are the bottleneck capacity for EB through movement, WB through movement, EB left-turn movement, and
WB left-turn movement, respectively.
The multi-objective optimization problem is addressed with a hierarchical method, also known as the lexicographic method ( Miettinen, 1999).
Generally, the through movements are synchronized controlled and considered more essential. Without loss of generality, we assume EB traffic
is prior to WB traffic, and then the objective functions in (1) are in the order of importance where f 1 is the most important and f 4 the least
important.
Constraints: In total, there are five sets of constraints:
(1) First, to simplify our discussion, we set the side street green splits and number of lanes unchanged in the
new CFI-Lite design. In this way, all the additional green time (if any) will be allocated to the major arterial.
(
Njk j
nk ; j 3; 4; 7; 8; and k i 1; i; i 1 2
Gk gk
j
j

Note this paper will not discuss the trade-off of the green time between the major and the side streets since we only aim to search for a
Pareto improvement. The overall performance of the CFI-Lite design could be further improved if green time for side streets is
changeable. This is left as a future research topic.
(2) Second, according to the CFI-Lite design, all the green times allotted to the arterial traffic are summed together at the bottleneck
intersection to serve left-turn and through movement simultaneously. Also, it is reasonable to assume that the number of lanes will be
kept unchanged. Therefore, we have:
24 W. Sun et al. / Transportation Research Part C 56 (2015) 18 33
( ni ; j 1; 2; 5; 6 3
Ni
Gi gi gi gi gi
j 1 2 5 6
j
j

Note that lost time is omitted here. It is easy to find that if lost time is considered, more benefit would be brought by the CFI-Lite design,
because the total number of phases is reduced.
(3) Third, the bottleneck capacity of the crucial streams has to be increased. Therefore, we have:

BPM P bPM ; P E; W; and M T; L 4

where b and B denote the bottleneck capacity in the conventional and CFI-Lite designs respectively. Superscripts E and W indicate
EB and WB traffic, and subscripts T and L indicate the through and left-turn traffic.
As vehicles traverse through three intersections, the bottleneck capacity is defined as the minimum value of the dis-charging flow rates of
three intersections. If we assume that these intersections use a common cycle length and the saturated discharging flow rate for each
single lane is the same, the maximum discharging flow rate for a movement can be represented by the product of green time and number
of lanes.
For the conventional intersections, the bottleneck capacity is simply constrained by the bottleneck intersection i.
Therefore, we have:

> bE gi ni
bL g5n5 5
> W i i

T
>

8 E
i
2 2
i
> b gn
< bLW
>

>
g1 in 1i

>
> T 6 6
>
>

But for the CFI-Lite intersections, since the geometry has been changed, the bottleneck capacity is impacted by all three intersections.
The following general equation can be used to calculate the bottleneck capacity for CFI-Lite intersections:
i
8 BT min G2 N2 ; G2N2; G2 N2
E i 1 i 1 i i
> i1 i1
>
>
6
>

> BE min q5
Gi 1Ni 1; Gi Ni
<
> L qi qi 1 5 5 5 5

> 5
5

> W 1 i i
> i 1 i i 1 i 1
> min G
>
BT 6

N6 ; G6 N6 ; G6 N6
>

W
>
>
i
>
>
>
q1 i 1 i 1 i i
> B min i G N ;GN
> L q q i 1 1 1 1 1
> 1 1
k i i 1 i i 1
:
where qj denotes the average flow rate for phase j at intersection k. q 5 q5 q1 and are all for left-turn move- q1
ments. Note the left movement has to first cross the street at upstream intersection and then turn left at intersection i.

From the above two equations, (4) can be rewritten as:


8min G2 i N2 ; G2N2; G2 N2 P g2n2
i 1 i 1 i i i1 i1 i i
>
>>
7
> q
i i
>min
>min
>
5
Gi 1Ni 1; Gi Ni G P g n
q
i
qi 1
N
5
; GN;G
5 5 N
5
Pgn
5 5

> 5 5

> 1 i 1 i i 1 i 1 i
> i i i

< 6 6 6 6 6 6 6 6
>
>
>
>
i
>
> >

>min
q1
i 1 G N ; GN
i 1 i 1 i i Pgn
i i
q 1 1 1 1 1 1
:

> qi
1
> 1

Clearly, the discharge capacity for both left-turn and through movements at the intersection i has been increased due to the specific
geometric design of CFI-Lite which allows left-turn and through traffic to move at the same time. Therefore, (7) can be further
expanded as the following two sets of equations which only consider upstream intersections:
8 gi ni
Gi 1Ni 1
< G2i 1N2i 1 P g2i n2i
>
>
6 6 P 6 6 ; for intersection i 1 8

>
qi

>
>
:
5 i 1 i 1 i i
> G N P g5n5
q5i q5i 1 5 5
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 25
8
Gi1Ni1 gi ni
< G2 i1
N2i1 P g2i n2i
; for intersection i 1 9

>
>
6 q
i 6 P 6 6

>
> 1 i 1 i 1 i i

> G N P g1n1
>
:

q1i q1i1 1 1
(4) Fourth, the capacity for other movements should not be reduced. Therefore, the following constraints need to be added
P
Gi 1Ni 1 gi 1ni 1
1 1 1 1
> 1 1
10

> G N Pg n
>

8 q5i 1 G i 1 N i 1 i 1 i 1

< i i g n
P
>> q5 q5 1 5
> i 1
5 5 i 1 i i
5
1 1
>
>
5 5 5
> 5
>
>
>

> i1 i1 i 1 i 1
qi1

>
:

qi1 G1 N1 P g1 n1
qi
(5) Last, the cycle length and total number of lanes for each approach of the upstream intersections should be
constant:

8 Nk Nk nk nk
k k
Nk Nk n n
< 2 5 2 5
;k i 1; i 1 11
> 6 1 6 1
>
>P
P P P
> 4 8 4 8
>
>
k k k k

>
> j1 g
j j5
g
j j1
G
j j5
G
j

Overall, the MO-MIP problem can be summarized as:


MO-MIP :Find fGjkg and fNjkg; to
max J f 1x BTE ; f 2x BTW ; f 3x BLE; f 4x BLW

s:t: 2; 3; and 811

4.2. Solution existence condition


The specifically designed constraints in the above MO-MIP make any feasible solution a Pareto improvement in the con-text of the CFI-Lite
design over the traditional design. So the solution existence condition for the MO-MIP also indicates whether the Pareto improvement exists.

Theoretically, the integer decision variables {N kj}, which indicate the number of lanes in the CFI design, have many com-binations of
different values. However, changing the number of lanes is strictly limited in reality. Considering the adaptabil-ity to different flow patterns, the
number of left-turn lanes is usually set one or two. Therefore, the number of combinations of {N kj} actually is very limited. In summary, we have
the following three possibilities:

(1) For the bottleneck intersection i, benefiting from the CFI-Lite design, left-turn traffic can share the green time with through traffic.
Consequently, one DLT lane for each approach is cost effective and enough to serve left-turn traffic at the bottleneck intersection.

(2) For phase 1 at intersection i 1 and phase 5 at intersection i + 1, the number of lanes can remain the same as in the traditional design
because traffic conditions are almost unchanged.
(3) For phase 5 at intersection i 1 and phase 1 at intersection i + 1, additional left-turn lanes may be required if green time is not enough for
the added left-turn volume. If so, the number of through lanes on the same approach must be reduced by one to maintain a constant total
number of lanes.

Based on above discussion, the integer variables can be reduced to two binary variables that indicate whether an additional left-turn lane is
needed. The following constraints regarding {Nkj} can be summarized
8 :
nkj;
k
> n y;
>
>j
>
>
> nk
>> j y ;

<
Nkj > nkj z;
>
>
>
nk z;
>> j
>
>
if j 1 and k i 1 ; 12
if j 5 and k i 1
if j 6 and k i 1
if j 2 and k i 1
else
where y 2 f0; 1g and z 2 f0; 1g are binary variables.
All four cases depending on the values of y and z are discussed in the following.
26 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

Case 1: y = 0 & z = 0: In this case, the number of lanes remains the same for every phase, i.e., N kj nkj.
For intersection i 1, phase 5 and phase 6 are critical conflicting phases. Pick out the relevant constraints from (8) and
(10) as follows:
i i 1 i 1 i i 1 i 1 i i
8 q5 q5
i i 1 G N i i G n Pg n
5 1

q5 iq5 5 q5 iq5 5 5 5 5
1 1
>q >
i
5
q5

q
i
5
1 Gi 1Ni 5
1
5 q q
i
q5
i 1 G ni
i 1
5
1
5
P 5
gi 1ni 1
5
13


> 5 5

>
<G 6
i1
N6i 1 G6i 1n6i 1 > g6in6i
> >
>
>

Assume that the initial green split setting {g kj} follows Websters principle (Webster, 1958), which is set proportional to the corresponding
volume of the phase, i.e.:
qi gi ni
5 5 5

> i
q5 q5
i1 i
g5 n5 g5 n5
i i1 i1

14
>
> qi 1 gi 1ni 1
8
>q i i n i 5 g i 51 i
g

5i 1 1
> q n
5 5 5

>
> 5 5 5

i i i
q g n
> 1 1 1
>q i qi ini gi ni 1
g

1 1


<

> 1
1 1
1 1 1
>
> i1 i1 i1
>
> :
q1 g1 n1
> i i i i i i 1
>

1 1

> 1 g 1 1

q q n g n
1 1 1

>
Then (13) is transformed to:
> G5i 1n5i 1 P g5in5i g5i 1n5i 1 n5 < g 5 g6 g n5
6 ni 1
G 5 n5 g5 g6 G6

1
>
8i i 1 i 1 i 1 i 1 i 1 i 1 i 1 i ni i 1 15
< 6

:
To satisfy both inequalities in (15), the necessary condition is:
ni
i1 16
n5i 1 g6i n5i 1 g5in5i > 0
6

g6 ni 1
6

Applying the same procedure to intersection i + 1, we can complete the solution existence condition.
gi gi n i > 0
i
8 n6
gi 1ni 1 ni 1
6 5 5 5 5 17
6i

> gi1 ni1 gi


6n
i 1
n
i 21 n gi ni > 0

:
i 1


> 2
2 1 1 1 1
< 2n

An example for the corresponding feasible solution is


> g5i 1 g5i; if j 5 and k i 1

g g;

G >
> i1 g i; if j 6 and k i 1
8 g6 5
> i 1 i
if j 1 and k i1 k k
g1;
> 18
if j 2 and k i1 ; Nj nj
> g2
> if j 1; 2; 5; 6 and k i
k
< 1 1
else
j
> > i 1 i

> k
>
> gj ;
>
i i
g1 g2
>
>
> ;
>
>

Case 2: y = 1 & z = 1: In this case, for both directions, the upstream left-turn movements are served with one additional lane, while the
corresponding through movements are served with one lane less.
For intersection i 1, besides that phase 5 and phase 6 are conflicting in green time, phase 5 and phase 2 are
conflicting in lane use. Similarly, the relevant constraints are picked out:

1 P g i ni i 1 i
Gi 1 ni 1 g n 1
> 5 5 5 5 5 5
< Gi 1 ni 1 1 < gi 1 gi 1 gi
n i
i6 ni 1 1
>
8 19


6
> >>
5 5 5 6 n6
1
5

: >
Gi2 1ni2 1 1 P gi2ni2
The necessary condition for inequalities in (19) to be satisfied simultaneously is:
gi gi ni gi
i i
> n n >0
i
6
ni 1 gi 1 gi 1 i
6

8 5 5 6
gi 1ni 1

>
6
n2i 1
15
g2 i
6 n6
1
5 5 6 n6
1
20

>
P
:

< i i 1

> n2 g2
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 27

Comparing to (16), one can find that (16) is relaxed in the first inequality in (20), because the expression in the round brackets is greater than
zero. At the same time, however, one more constraint for phase 2 is added. Actually, the first inequality in (20) is very likely to be satisfied in
real world. For example, if the number of lanes at consecutive intersections along an arterial is equal, and single left-turn lane is used, the first
inequality in (20) is written as
g 5i 1 g6i 1 g5i g6i g 6i 1 g6i > 0 21

Because intersection i is the bottleneck intersection that has heavy side street traffic, each of the two expressions in parentheses in (21) is greater
than zero. So (21) is always satisfied. Under this condition, the original necessary conditions
(20) could be reduced to
1 gi
ni 1 2 2

ni P gi 1 22
2 2

Comparing the solution existence condition at intersection i 1 in Case 1 and Case 2, we can find that if g i6 1 is relatively large, (16) is likely to be
satisfied; if gi2 1 is relatively large, (22) is likely to be satisfied. Consequently, we can further benefit from this flexibility by using changeable
lane. That is, set y = 0 when EB traffic is heavy, set y = 1 when WB traffic is heavy. Applying the same procedure to intersection i + 1, we can
complete the solution existence condition.
ni
1 g
i
>
gj k
1
> 2 2
P
8
g
g n g n gn g g i 1 >0

i
i 1

n2 g2i 1
>
>
i i
> i 1 i 1 i n6 i 1 i i i 1 i 1 i n6
6 5 5 5 5 5 6
>

> 6n 6 6n 6
>

>
i

i 1 1 g
> n
>
>
6 6
> g n
i 1g >0

i 1 n gn g g g
i 1
> P
i


<
>
n6 g6
>
>
> i i
>
> i 1 i 1 i n2 i 1 i i i 1 i 1 i n2

>

> 2 2
:

1 1 1 1 2
>2 2n 1 2n

An example for the corresponding feasible solution is


1; if j 5 and k i 1
ni 1
G
;N > 5 1; if j 1 and k i 1
k
g1
i
g2;
i if j 1; 2; 5; 6 and k i k 8n2i 1 > i 1
1; if j 2 and k i 1
>

j
(
k j
> 1
gj ; else >>

>
i 1
>n 1; if j 6 and k i 1
< 6
>

>
>
:

> else
njk;
Case 3: y = 0 & z = 1: Similarly as discussed above, we can easily find a solution
gi ;
if j 5 and k i 1 1; if j 1 and k i 1
gi 1 ni1
5 5
>g if j 1 ; 2; 5; 6 and k i
k
8 g6 1
1 g2;
g5; if j 6 and k i 1
k >
i 1
> i i k
Gj > i i ; Nj 8 n6 1; if j 6 and k i 1
>
> :
<
>
<>
n; else
:

> else >j


gjk;
which is feasible on the following conditions:
gi gi ni > 0
i
8 i n6
gi 1ni 1 i
ni 1
6 5 6 1 5 5 5
i 1 n6
> 6
>
> n6 1
> g6
i
> n P i 1
> 6 g
>
>
>
>
< ni ni
>
:
gi1ni1 gi i
2
ni1 gi ni gi1 gi1 gi i
2
>0

>2
1
2 n21 11 1 1 22 n21

Case 4: y = 1 & z = 0: Similarly as discussed above, we can easily find a solution


gi ; if j 1 and k i 1 1; if j 5 and k i 1
gi1 ni 1
1 1
> if j 1 ; 2; 5; 6 and k i
k
8 g2
g1 g2;
g1 ; if j 2 and k i 1
k >
i
G
> i 1 i k
j > i i ; Nj 8 n2 1 1; if j 2 and k i 1
>
> :
<
>
<
>
n; else
:
23

24

25

26

27
28 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

which is feasible with the following conditions:


i1 i
> n2 1 g2
P
g n g i 1 n gn g g g i 1 >0 28

i i i
8 n2
g2i 1
>
>
>
>
i 1 i 1 i n6 i 1 i i
i 1
i 1 i n6
>
6 6

6 5 6n 5 5 5 5 6 6 n
>
>
>
>
<
>
gi1ni1 gi n i
i2 ni1 gi ni > 0
:

1 1 1
>2 2 n2
1
1

4.3. Solving algorithm

Finally, we offer a procedure to solve the proposed MO-MIP. Generally, solving a MO-MIP is not an easy work. However, the MO-MIP
discussed in this paper can be simplified and solved very quickly.
First, the hierarchical multi-objective optimization problem could be solved by solving a sequence of single-
objective
optimization problem. Considering the objectives in the MO-MIP, maximization of BE and BW is not conflicting, and
T T
E W
maximization of BL and BL is not conflicting either. So (1) is equivalent to
max F2x BLE BLW 29

s:t: F1x BTE BTW F1


where F1(x) is the principle objective function, F2(x) is the secondary objective function, F1 is the optimal (maximum) value of F1(x).

Second, as discussed in Section 4.2, the values of integer variables in the MO-MIP are restricted to four cases. Therefore, we can solve one
by one the four subsequent problems without integer decision variable, and select the best one from the four.

Lastly, all the constraints and the objective functions of the MO-MIP are linear. We can apply simplex method to solve these linear functions.

The solving algorithm is expressed as follows:

Step 1: determine the values of integer variables using a decision tree model as shown in Fig. 4.
Step 2: if no solution exist, end the algorithm; otherwise, for each pair of feasible values of y and z, solve the following linear programming
(LP) problem by simplex method.

LP1m : Find fGkjg; to


max J F1x BET BWT
s:t: 2; 3; and 811

y; z& ym; zm&


where LP1m is the mth LP problem to maximize the primary objective, m = 1, 2, . . . , M; M 6 4, ym and zm are the mth feasible values of y and
z, respectively.
Step 3: obtain the optimal value of F1(x), F1, by selecting the best results from the above M LP problems, and then solve the subsequent LP
problem (see the following) to optimize the secondary objective.

Fig. 4. The decision tree model for determining the values of y and z.
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 29

Fig. 5. The CFI-Lite design for consecutive bottleneck intersections.

LP2 : Find fGkjg; to

max J F2x BEL BWL


s:t: 2; 3; and 811
y; z& y ; z &
F 1x F1


where LP2 is the LP problem to maximize the secondary objective, y and z are the optimal values of y and z that maximize the primary
objective, respectively.

5. CFI-Lite design for consecutive bottlenecks

In some of the urban arterials, bottleneck intersections are located consecutively. Here we briefly introduce how the CFI-Lite design is
expanded to such scenarios.
Fig. 5 shows an example how the CFI-Lite design is applied to two consecutive bottlenecks. Arterial left-turn movements consist of 3 parts in
this system. Take EB traffic as an example: at the upstream intersection, all vehicles that intend to turn left at bottleneck intersections B1 and B2
should turn left to the DLT lane; at bottleneck intersection B1, some of the vehicles in the DLT lane turn left to side street, and some go through
to downstream DLT lane; and at bottleneck intersection B2, all vehicles in the DLT lane turn left to side street.

In comparison to the original design shown in Fig. 1, the differences are:

(1) At bottleneck intersection B1, WB right turn vehicles cannot pass the intersection when arterial is given green time. Instead, they can
pass the intersection when side street is given green time.
(2) At bottleneck intersection B2, EB right turn vehicles cannot pass the intersection when arterial is given green time. Instead, they can pass
the intersection when side street is given green time.
(3) At upstream intersections, the traffic demand turning left to DLT lane is much heavier. Therefore, one more left-turn lane might be
needed to meet the solution existence condition in such a scenario.
(4) Signal control for a system of consecutive CFI-Lite intersections is similar as we discussed in the previous section. However, the
existence of Pareto improvement solutions must be carefully checked. That is because new constraints are added to the CFI-Lite design
for consecutive bottlenecks. And the larger the number of consecutive bottlenecks is, the more constraints need to be considered.

6. Case study

In this case study, we use the field data from a real-world network illustrated in Fig. 6(a). The network is a segment of Foothill Blvd., a major
arterial in the city of Los Angeles, CA. It consists of four signalized intersections located at Emerald, D St., Fruit-White, and Towne Center from
west to east, respectively. The arterial is controlled by an actuated coordinated system. If a non-synchronized phase does not use all of the
allocated time, then the extra time will be given to the synchro-nized phases 2&6.

Among the four intersections, Int. Fruit-White is a critical intersection for the arterial with almost equal amounts of main-line and cross-street
traffic. Split phasing is used at Int. D St., where phase 7 (all southbound traffic) and phase 8 (all
30 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

Foothill
blvd

900 feet 1200 feet 1100 feet

(a) 14s 18s 17s

Emerald D st Fruit-White Towne center

(b)

DLT lane DLT lane


Foothill

blvd DLT lane DLT lane

(c) Emerald D st Fruit-White Towne center

Fig. 6. The test arterial: Foothill Blvd, Los Angeles, CA: (a) current conventional design; (b) phase denomination and signal timing plan; (c) CFI-Lite design.

northbound traffic) together occupy a large amount of green time. Emerald and Towne center are minor intersections where the green times
allocated to cross street are relatively few. So, Int. Fruit-White and Int. D St. are the bottleneck intersections.
Signal timing data collected during evening peak from 15:00 p.m. to 19:00 p.m. on five workdays, December 26, 2013, are used in this case
study. The phase denomination and signal settings for the time period of analysis are given in Fig. 6(b). To investigate the bottleneck capacity for
the system, signal cycles are combined as groups of four. In each cycle group, one cycle from each of the four intersections is selected, and the
difference of local cycle start times for adjacent intersections must be exactly the same as the relative offsets. Cycles excluded in any cycle group
(e.g., transition cycles) are omitted here. In this case, 408 cycle groups in all were studied.

The corresponding CFI-Lite design for the network is shown in Fig. 6(c). In accordance with Section 5, DLT lanes are uti-lized at the median
two bottlenecks, Int. D St. and Int. Fruit-White. Signal control for such a design was determined following the similar procedure we proposed in
Section 4. For the bottlenecks, the green times for these two intersections are the sum of green times initially allocated to phase 1 and phase 2 in
the conventional design. For upstream intersections, we assume that all the EB left-turn vehicles (phase 5) at the bottlenecks come from EB
through vehicles (phase 2) at Int. Emerald, and all the WB left-turn vehicles (phase 1) at the bottlenecks come from WB through vehicles (phase
6) at Int. Towne Center. Note that this assumption is conservative, since some of the left-turn vehicles may come from cross street in reality. As a
result, the green time initially allocated for EB left-turn traffic are requested to be added to phase 5 at Int. Emerald, and WB left-turn traffic
added to phase 1 at Int. Towne Center. To handle the heavy left-turn movements, the solution suggested in Case 2 in Section 4.2 was used. One
through lane for phase 2 is changed to left-turn lane for phase 5 at Int. Emerald, and one through lane for phase 6 is changed to left-turn lane for
phase 1 at Int. Towne Center. We keep the green times for upstream left-turn to DLT lanes just enough to accommodate the total left-turn demand
so as to maximize the through capacity. Note again the green times for side streets are kept unchanged.

To evaluate the performance, here we compare the values of bottleneck capacity for synchronized phases.
Considering the change of lanes, the effective green time is calculated as the green time multiples by the lane
change ratio, where the lane change ratio is defined as the number of lanes in the CFI-Lite design divided by the
number of lanes in the conventional design. The lane change ratios are: 2, for phase 5 at Int. Emerald and phase 1
at Int. Towne Center; 2/3, for phase 2 at Int.
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 31

Fig. 7. Comparisons of effective green times: (a) phase 2; (b) phase 6.

Emerald; 1/2, for phase 6 at Int. Towne Center; and 1, for others. The comparisons of average effective green time for the synchronized phases
are shown in Fig. 7. By applying a CFI-Lite design, the bottleneck effective green time for phase 2 is increased by 48.8% from 32.6 s to 48.5 s,
and the bottleneck effective green time for phase 6 is increased by 23.1% from 32.1 s to 39.5 s. This shows a tremendous improvement.

We further check the cycle groups one by one. The distribution of the improved capacity in terms of effective green time is shown in Fig. 8.
The bottleneck capacity for phase 2 is reduced in only 7 out of 408 cases, demonstrating a 98.3% possibility of improvement (Fig. 8(a)). For
phase 6, however, the CFI-Lite design fails to outperform the conventional design in 38 out of 408 cases ( Fig. 8(b)). The number of through
lanes accounts for the difference. Only two through lanes are available for phase 6 at Int. Towne Center in the conventional design, and one of
them is changed to serve increased left turn traffic in the CFI-Lite design, leaving only one through lane serving phase 6. But the possibility of
improvement is still as high as 90.6%. The bottleneck capacity for WB through traffic can be further improved by changing the right turn lane to
through lane if right turn demand is not very high.

7. Concluding remarks

In this paper, we propose a simplified continuous flow intersections (called CFI-Lite) design for oversaturated bottleneck intersections.
Similar to the CFI design, left-turn traffic is allocated in advance to the displaced left-turn lanes at upstream intersections, so that they can pass
the bottleneck intersection without conflicting the through traffic. But the CFI-Lite design does not need installation of sub-intersections, so it is
ideal for urban arterials where intersections are closely spaced. Road works, although not heavy, are required to rebuild conventional
intersections to CFI-Lite ones. Overall, it is a cost-effective way of relieving oversaturated traffic conditions in urban areas.

The CFI-Lite design provides a good inspiration to relieve oversaturation. Intersections adjacent to the bottleneck inter-sections are
coordinated not only to gate traffic, but also to sort traffic so that they can pass the bottleneck intersection more effectively. In this way, the CFI-
Lite intersection provides superior capacity to the original CFI, because the redundant green time at less saturated upstream intersections are
utilized to address the conflict between left-turn and through movements.
To operate the signals in CFI-Lite intersections, this research further formulates the signal control at CFI-Lite intersections as a multi-
objective mixed integer programming problem and seeks for Pareto capacity improvements. The most important
32 W. Sun et al. / Transportation Research Part C 56 (2015) 1833

Fig. 8. Distribution of improved capacity: (a) phase 2; (b) phase 6.

consideration is to provide sufficient green time for left-turn movements at the intersection upstream of the bottleneck; so that both the original
left-turn traffic and the newly arisen traffic turning left to the DLT lane could be accommodated. The preliminary result in this paper shows that
the Pareto capacity improvement for all the traffic streams could be achieved under effortless conditions.

Link length is an essential parameter that influences how much benefit a CFI-Lite design can gain. If the link length is too long, the adverse
effects caused by the DLT lane to bus stops and mid-street sources/sinks may become unacceptable. On the other hand, if the link length is too
short, the risk of spillover in the DLT lane should be carefully considered. With the only available real-world data in hand, we tested the Foothill
Blvd that has link length around 1000 feet in the case study. We will try to get more field data and see whether the CFI-Lite design can help
improve capacity for arterials with various link lengths.

Progression between signals is another critical issue that determines the bottleneck capacity. Thus, the offset and phasing sequence in a CFI-
Lite design should be set up accordingly. But this paper mainly focuses on introducing the basic concept of the CFI-Lite design and its potential
benefits. By now, we simply assume that the phase sequence and offset in the CFI-Lite design remain the same as in the conventional design.
Optimization of phasing sequence and offset is one of important future research topics.

Lastly, the detailed design standard or guidance of the CFI-Lite design such as lane channelization, pavement markings and guidance signs,
could significantly reduce any potential inconvenience or confusion for drivers and passengers brought by the new CFI-Lite. It will be addressed
in future research.

Acknowledgements

This research was funded partially by the National Science Foundation of China under Grants #51328801 and #51278021, and the
Fundamental Research Funds for the Central Universities. The views are those of the authors alone.

References

Abramson, P., Bergren, C., Goldblatt, R., 1995. Human factors study of the continuous flow intersection at the dowling college NAT center. In: Presented at 4th Annual
Symposium on Intermodal Transportation, Oakdale, NY, p. 2.
Abu-Lebdeh, G., Benekohal, R.F., 2000. Genetic algorithms for traffic signal control and queue management of oversaturated two-way arterials. Transport.
Res. Rec. 1727, 6167.
W. Sun et al. / Transportation Research Part C 56 (2015) 1833 33

Autey, J., Sayed, T., El Esawey, M., 2013. Operational performance comparison of four unconventional intersection designs using micro-simulation. J. Adv.
Transport. 47 (5), 536552.
Bared, J., 2009. Displaced Left-Turn Intersection. Publication FHWA-HRT-09-055. FHWA, US Department of Transportation.
Carroll, D.H., Lahusen, D., 2013. Operational effects of continuous flow intersection geometrics a deterministic model. Transport. Res. Rec. 2348, 111.
Chang, T.H., Lin, J.T., 2000. Optimal signal timing for an oversaturated intersection. Transport. Res. Part B: Methodol. 34 (6), 471491.
Daganzo, C.F., 2007. Urban gridlock: macroscopic modeling and mitigation approaches. Transport. Res. Part B: Methodol. 41 (1), 4962.
El Esawey, M., Sayed, T., 2007. Comparison of two unconventional intersection schemes crossover displaced left-turn and upstream signalized crossover intersections.
Transport. Res. Rec. 2023, 1019.
El Esawey, M., Sayed, T., 2013. Analysis of unconventional arterial intersection designs (UAIDs): state-of-the-art methodologies and future research
directions. Transport. A: Transport Sci. 9 (10), 860895.
Gazis, D.C., 1964. Optimum control of a system of oversaturated intersections. Oper. Res. 12 (6), 815831.
Goldblatt, R., Mier, F., Friedman, J., 1994. Continuous flow intersection. Inst. Transport. Eng. J. 64 (7), 3542.
Hu, H., Wu, X., Liu, H.X., 2013. Managing oversaturated signalized arterials: a maximum flow based approach. Transport. Res. Part C: Emerg. Technol. 36 (1), 196211.

Hughes, W., Jagannathan R., Sengupta, D., Hummer, J.E., 2010. Alternative Intersections/Interchanges: Informational Report (AIIR). Publication FHWA-HRT-09-060. FHWA,
US Department of Transportation.
Hummer, J.E., Reid, J.D., 2000. Unconventional left-turn alternatives for urban and suburban arterials: an update. In: Transportation Research Circular E-
C019: Urban Street Symposium Conference Proceedings, 2830 June, Dallas, TX.
Jagannathan, R., Bared, J.G., 2005. Design and performance analysis of pedestrian crossing facilities for continuous flow intersections. Transport. Res. Rec.
1939, 133144.
Lieberman, E.B., Chang, J.N., Prassas, E.S., 2000. Formulation of real-time control policy for oversaturated arterials. Transport. Res. Rec. 1727, 7788.
Liu, Y., Chang, G., 2011. An arterial signal optimization model for intersections experiencing queue spillback and lane blockage. Transport. Res. Part C:
Emerg. Technol. 19 (1), 130144.
Michalopoulos, P.G., Stephanopoulos, G., 1977a. Oversaturated signal systems with queue length constraintsI: Single intersection. Transport. Res. 11 (6), 413421.

Michalopoulos, P.G., Stephanopoulos, G., 1977b. Oversaturated signal systems with queue length constraintsII: Systems of intersections. Transport. Res. 11
(6), 423428.
Miettinen, K., 1999. Nonlinear Multiobjective Optimization. Springer.
National Cooperative Highway Research Program (NCHRP), 2012, NCHRP 03-09 Report: Operation of Traffic Signal Systems in Oversaturated Conditions.
<http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_w202v2.pdf> (accessed 22.10.14).
National Transportation Operations Coalition (NTOC), 2011, 2011 Traffic Signal Operations Self Assessment, NTOC. <http://www.ite.org/selfassessment/> (accessed 22.10.14).

Park, B., Messer, C.J., Urbanik, T., 2000. Enhanced genetic algorithm for signal-timing optimization of oversaturated intersections. Transport. Res. Rec. 1727, 3241.

Reid, J.D., Hummer, J.E., 2001. Travel time comparisons between seven unconventional arterial intersection designs. Transport. Res. Rec. 1751, 5666.
Webster, F.V., 1958. Traffic Signal Settings. Road Research Technical Paper, No.39, Great Britain Road Research Laboratory, London.
Wu, X., Liu, H., Getttman, D., 2010. Identification of oversaturated intersections using high-resolution traffic signal data. Transport. Res. Part C: Emerg.
Technol. 18 (4), 626638.
Wu, X., Juarez, D., Jia, X., 2014. Optimal signal timing models for the FHWA and Mexico 4-legged continuous flow intersections. In: Presented at the Transportation Research
Board 93rd Annual Meeting, January 1216, Washington, D.C.
Xuan, Y., Daganzo, C.F., Cassidy, M.J., 2011. Increasing the capacity of signalized intersections with separate left turn phases. Transport. Res. Part B:
Methodol. 45 (5), 769781.
Yang, X.F., Chang, G.L., Rahwanji, S., Lu, Y., 2013. Development of planning-stage models for analyzing continuous flow intersections. J. Transport. Eng. 139 (11), 11241132.