A Two-level Traffic Light Control Strategy for

Preventing Incident-based Urban Traffic Congestion

Liang Qi, Student Member, IEEE, MengChu Zhou, Fellow, IEEE, and WenJing Luan, Student Member, IEEE

Abstract—This work designs a two-level strategy at signalized safe crossing of conflicting streams of vehicles and pedestrians
traffic intersections for preventing incident-based urban traffic and lead to efficient network operations [8]. Much work has
congestion by adopting additional traffic warning lights. The been done to develop various traffic signal strategies for
first-level one is a ban signal strategy that is used to stop the promoting traffic efficiency. They are mainly categorized as
traffic flow driving towards some directions; and the second-level
fixed-timed, traffic responsive, and predictive control
one is a warning signal strategy that is to give traffic flow a
recommendation of not driving to some directions. As a visual strategies. The first one is widely adopted in most current
and mathematical formalism for modeling discrete-event urban traffic systems due to their easy implementation and low
dynamic systems, timed Petri nets (TPNs) are utilized to describe management cost. However, its drawbacks are that the
the cooperation between traffic lights and warning lights, and resulting settings are based on historical rather than real-time
then verify their correctness. A two-way rectangular grid data. The second one, such as SCOOT [9] and SCATS [10], is
network is modeled via a cell transmission model. The based on measured current traffic states and have been
effectiveness of the proposed two-level strategy is evaluated effectively used in many cities around the world [8]. In [11]
through simulations in the grid network. The results reveal the Dotoli et al. propose a discrete-time-model-based signal
influences of some major parameters such as the route-changing
timing plan and show its effectiveness in dealing with severe
rates of vehicles, operation time interval of the proposed strategy,
and traffic density of the traffic network on a congestion traffic congestion. The third one is a model-based optimization
dissipation process. The results can be used to improve the state control strategy that can forecast the future traffic behavior of
of the art in preventing urban road traffic congestion caused by the network based on traffic-forecasting models [12]. Traffic
incidents. light strategies can be designed for single-intersection control
 and network-wide control. For a single signalized intersection,
an optimal traffic light switching scheme can be computed
Index Terms—Timed Petri net (TPN), cell transmission model efficiently to minimize the queue lengths described by a model
(CTM), traffic light control, emergency strategy, congestion
[13]. Recently, there are some researches about the
formation and dissipation, traffic incident.
network-wide traffic control [14]-[22]. Aboudolas et al. [14],
[15] propose a real-time signal control methodology based on
a store-and-forward modeling paradigm by analytically
I. INTRODUCTION
solving a quadratic-programming problem that aims at

W ith a growing number of vehicles in many countries,

traffic congestion becomes a serious problem and has
spread like a plague from major metropolitan areas to
minimizing and balancing the link queues. Geroliminis et al.
[18]-[22] obtain many theoretic results of a macroscopic
fundamental diagram that can be utilized to establish efficient
numerous small to mid-size cities. It increases human travel and elegant strategies to control network flows.
time, fuel consumption and air pollution. The current research
Most of the above theories and strategies are suitable for
[1] finds that a congestion can be caused by one of the
non-saturated and stable traffic conditions. However, changing
following three reasons: 1) a temporary obstruction within a
environmental conditions in a non-predictable way such as an
traffic network such as an incident, 2) a permanent capacity
incident may lead to an invalidation of the above traffic light
constraint, and 3) a stochastic fluctuation in the demand. Note
control strategies, and causes traffic jam or unexpected
that the incident-based congestion is a traffic jam that can
congestion. To solve this problem, route guidance and driver
cause large-scale congestion to the upstream traffic flow if the
information systems can be employed to improve the network
incident is not cleared in a timely fashion. The rapid
efficiency via direct or indirect recommendation of alternative
developments in communications and computing technologies
routes [8]. The real-time traffic information can be sent to
promote several advanced traffic control strategies to be
drivers through two kinds of devices: in-car [17] or road-side
proposed to deal with these problems [2]-[8].
devices [2]. The first one, such as radio and GPS-navigator,
Among those strategies, the traffic light control at road helps drivers make sensible routing decisions at bifurcation
intersections is regarded as a major strategy to guarantee the nodes of the network. However, drivers who are familiar with
the traffic conditions in a network may not use such devices
Manuscript received January 10, 2016; revised May 13 and July 20, 2016; and optimize their individual routes based on their past
accepted October 28, 2016. This work was supported in part by the National experience. The second one includes electronic signs
Natural Science Foundation of China under Grant 61202028, the U.S.
displayed on roadways which provide present travel time,
National Science Foundation under Grant CMMI-1162482 and the
Scholarship from the China Scholarship Council. The Associate Editor for this traffic congestion and special events. For example,
paper was ***. (Corresponding author: MengChu Zhou.) Orlando-Orange County Expressway Authority in Florida
L. Qi and W. Luan are with the School of Electronics and Information added 29 fixed dynamic message signs (DMS) to their toll
Engineering, Tongji University, Shanghai 201804, China (e-mail:
[email protected], [email protected]).
road network between 2006 and 2008 for traffic diversion [2].
M. Zhou is with the Department of Electrical and Computer Engineering, Al-Deek et al. discuss the role of traffic information provided
New Jersey Institute of Technology, Newark, NJ 07102 USA (e-mail: by DMS [5]. Though such devices can provide comprehensive
[email protected]). information, they usually have great management cost and
increase the drivers’ decision-making complexity. Currently modeled with an extended CTM, and then evaluates the
DMS is mainly installed on some arterial highways. effectiveness of the proposed two-level strategy via simulation.
Section V concludes this paper.
So far incident-based urban traffic congestion is mostly
controlled and prevented through traffic flow diversion with
the help of the traffic police. Ban signals are usually adopted
[23], [24] so as to notify road users of a ban situation that II. TRAFFIC LIGHT CONTROL
might not be readily apparent otherwise. In many countries, This section aims to describe the traffic light phase
ban signs take the shape of an equilateral triangle with a white transitions at signalized traffic intersections, and build the
background and a thick red border. These strategies are normal traffic light control strategy based on TPNs.
unfortunately labor-intensive, inflexible and non-intelligent.
This work proposes to adopt additional traffic warning lights
for sending incident or congestion warnings. These lights can A. Traffic lights
be easily installed at signalized intersections and are divided We assume that each of the four directions north (n), east (e),
into two kinds: ban signals and warning ones. Their use and south (s), and west (w) deploys a traffic light set. As shown in
cooperation with traditional traffic lights at road intersections Fig. 1, each set including 9 lights that have the capacity of
yield a novel and more flexible two-level emergency strategy sending seven kinds of traffic light signals: a red one (R), three
for preventing traffic congestion caused by incidents. This yellow ones with a left turn arrow (YL), a straight arrow (YS),
work proposes to utilize timed Petri nets (TPNs) [25], [26] for and a right turn arrow (YR), and three green ones with a left
the design of traffic light control strategies: normal one using turn arrow (GL), a straight arrow (GS), and a right turn arrow
normal traffic lights and emergency one using warning lights. (GR), respectively. In our work we deploy a four-phase traffic
Through these TPN-based formal models, it successfully light rule to direct the traffic flow at a signalized intersection as
verifies the correctness of the control systems. The strategy shown in Fig. 2. The phase transition forms a cycle, i.e., from
can consequently be implemented directly with hardware and phases a to d, and then back to a. For convenience, Xn, Ye, Zs,
software co-design. Note that the reason to adopt PNs is their and Ww are denoted as traffic lights along directions n, e, s, and
graphical and mathematical representation ability to describe w, respectively, where X, Y, Z, W{R, YS, YL, YR, GS, GL,
well the cooperation of traffic lights [27]-[34]. Recently, some GR}. The detailed operations are described next and modeled
studies use PNs to model the highway traffic [35] and by a TPN model. For convenience, in the following discussion,
signalized traffic intersections[36]-[39] in a microscopic way. let χ{n, e, s, w}, ξ{n, s}, and ζ{e, w}.
In order to better understand the dissipation of traffic
congestion, many researchers have explored its formation. B. Traffic light control model
They perform that through simulation by adopting an accident
as an obstruction in their traffic network model. Some Traffic light control systems with traffic light phase
simulation models are proposed in [1], [23], which concentrate transitions are regarded as discrete event systems. In order to
on a holistic view of traffic jam formation due to various ensure such systems to be correctly implemented by the
incidents. Wright and Roberg propose an incident-based jam co-design of hardware and software, we first use TPN as a
growth model in [1] and discuss the impact of the length of the formal model to perform its design.
channelized part of roads and stopline width assignment on PNs [25], [26], [40]-[43] are a type of bipartite directed
jam formation. Roberg et al. develop several alternative graphs with four types of objects, i.e., places, transitions,
strategies in [40] to prevent the gridlock of a network and directed arcs connecting places to transitions and transitions to
dissipate traffic jams once they have formed. Long et al. [24] places, and inhibitor arcs from places to transitions.
have extended a cell transmission model (CTM) and applied it Deterministic delays associated with PN transitions lead to
to simulate incident-based jam propagation in two-way
rectangular grid networks. They also propose control strategies
for dispersing incident-based traffic jam and evaluate their
efficiency. Compared with the traffic flow models utilized in
one-way networks [23], [40], CTM can simulate network-wide
traffic flow in a more realistic manner. Furthermore, two-way Fig. 1. A traffic light set.
roads are more commonly found in urban traffic networks than
one-way ones. In this work, the proposed strategy is evaluated
through a simulation study in a two-way rectangular grid
network modeled via an extended CTM. The impact of
route-changing rate of vehicles owing to traffic warning lights
is investigated and the effectiveness of the proposed strategy is Phase a n Phase b
evaluated. Besides, because of the real-time requirement of the w e
operation and the hard-measurability of an accident, the
s
strategy response time is investigated.
The rest of this paper is organized as follows. Section II
presents a signalized intersection and traffic light control
model via TPNs. Section III gives a two-level traffic light Phase d Phase c
control strategy and performs its correctness verification.
Section IV presents a two-way rectangular grid network Fig. 2. Four phase transitions of the traffic control system.
TPNs. First, some basic notations are presented: is a real obtains a token. At this time, tξ5 and tξ6 are enabled. Given
number set; + is the set of positive real numbers; and ={0, 1, τ(tξ5)=τ(tξ6)=0, both transitions fire immediately resulting that
2, } is a natural number set. Then we give the detailed each of pξ4 and pξ5 obtain a token and each of pi2, pi3, pi6, and pi7
definition of TPNs. obtains a token, thus enabling tξ7 and tξ8. It denotes that the
lights change from Rξ to GRξ and GSξ and the duration time of
Definition 1: A TPN is an 8-tuple =(P, T, F, W, C, H, M0, τ), GRξ and GSξ are 15 seconds because (tξ7)=(tξ8)=15. tξ7 and tξ8
where fire at G=20 such that tokens are moved from pξ4 and pξ5 to pξ6
1) P is a finite set of places; and pξ7, respectively. This denotes that the lights change from
GRξ and GSξ to YRξ and YSξ and the duration time of YRξ and
2) T is a finite set of transitions with PT and PT=; YSξ are 3 seconds. After that, tξ9 and tξ10 fire such that tokens
3) F(PT)(TP) is a set of directed arcs; are moved from pξ6 and pξ7 to pξ10 and pξ11, respectively. The
immediate transition tξ2 is enabled and fires by removing the
4) HPT is a set of inhibitor arcs from P to T with tokens in pξ10 and pξ11 and depositing a token to pξ3. This means
HF=; that the traffic lights GLξ are turning on at G=21 and the
duration time of GLξ is 10 seconds since (tξ3)=10. Then the
5) W: (PT)(TP)→ is a weighting function, where
light changes from GLξ to Yξ at G=31, i.e., tξ3 fires with a token
W(f)>0 if fFH and otherwise W(f)=0;
removed from pξ3 in pξ2. At the same time, a token is filled into
6) M: P→ is a marking function, where pP, M(p) is the each of pi1, pi4, pi5, and pi8. Then tξ4 with (tξ4)=2 and tζ1 with
number of tokens in p, and M0 denotes the initial marking; (tζ1)=4 are enabled and fire at G=33 and G=35, respectively. It
means that the duration time of YLξ is 2 seconds. Note that, tξ1
7) K: P→ is a capacity function, where pP, K(p) is the is not enabled at G=35 because each of 1-capacitied places pi2,
maximum number of tokens that p can hold;
pi3, pi6 and pi7 contains a token. Then the traffic lights
+
8) τ: T→ is the firing time of transitions. successively turn to GSζ and GRζ, YSζ and YRζ, GLζ, YLζ, and
back to their initial states when G=63. Then tξ1 fires at G=65,
In TPNs, each transition t is attached with a fixed enabling and the same transition firing sequence is executed repeatedly.
time duration τ(t). If t becomes enabled, then it fires after time For convenience, we denote markings of the TPN as the states
duration τ(t). If τ(t)≠0, t is called a deterministic transition. If of the lights, and have M0=(Rξ, Rζ), M1=(GSξ, GRξ, Rζ), M2=(YSξ,
τ(t)=0, t is called an immediate transition and fires at the YRξ, Rζ), M3=(GLξ, Rζ), M4=(YLξ, Rζ), M5=(Rξ, Rζ), M6=(GSζ,
moment when it becomes enabled. Graphically, immediate and GRζ, Rξ), M7=(YSζ, YRζ, Rξ), M8=(GLζ, Rξ), and M9=(YLζ, Rξ).
deterministic transitions are drawn by thin and thick black bars, The reachability graph of the TPN model is shown in Fig. 4.
respectively. We adopt an inhibitor arc to design a compact PN. Since this graph containing all transitions is a finite circle, the
It links place p to t with a small circle attached to t. It prevents t TPN model is live and reversible. The time necessary to
from firing as long as p is marked if (p, t)H. In the model in complete a cycle from marking Mi, i + to itself is 60 seconds
this paper, we specify that each arc has a weight of 1 and the except that from M0 to M0 for the first time which is 63 seconds.
capacity of each place is 1 (i.e., the net is 1-bounded [26]). For The proposed model makes sure that the order of the traffic
convenience, let G denote the global time and G0 the initial lights forms a cycle. It directs the traffic light phases operating
time. When a TPN model is built, we can construct its in a cycle “a→b→c→d→a”. Note that M1 and M5 are not same
reachability graph by following the firing rules [26]. The because token counts in pi1-p i8 are different.
graph size can be reduced by attaching to each arc a sequence
of transitions containing only one deterministic transition.
Note that if more than one deterministic transition fires at the pw10 tw2 pw11
same time, they are attached to the same corresponding arc tw9 tw10
and are put in a parenthesis. Now we use TPNs to construct a pw6 pw7
pn8 tn5 pn4 tn7 pn6 tn9 pn10
normal traffic light control strategy. tn1 tn2
tw7 tw8
In our discussion, the traffic light control policy for a tn4 tn3 pn9 tn6 pn5 tn8 pn7 tn10 pn11
pw4 pw5
signalized intersection satisfies the following statements [31]: pn1 pn2 pn3 pe1
tw5 tw6 te4
 The order of traffic lights forms a complete cycle; pw8 pw9 i
p1 i
p7 pe2
 The policy starts when all red lights are on; tw1 pw3 pi2 i te3
 A yellow duration is adopted during the light transition p8
from green light to red one; and tw3 pi6 te1
pi4
 To insure traffic safety, all red lights are on when the pw2
i
p3 pi5 pe3 pe8 pe9
light transition is completed. tw4 te6 te5
Fig. 3 presents the TPN model of a normal traffic light pw1 ps3 ps2 ps1
pe5 pe4
strategy. Table I describes the meanings of places and ps11 ts10 ps7 ts8 ps5 ts6 ps9 ts3 ts4
transitions with their enabling durations. Note that we assume te8 te7
ts2 ts1
such time durations for an illustration purpose. They can be pe7 pe6
ps10 ts9 ps6 ts7 ps4 ts5 ps8
changed to user-desired values based on a real-time traffic te10 te9
condition. This will not affect the control structure of the
pe10 pe11
strategy. Let G0=0 at the initial state of the TPN, i.e., only pχ1 te2
contains a token at G0. tξ1 is enabled and fires after 4 seconds
such that a token is moved from pξ1 and each of pξ8 and pξ9 Fig. 3. TPN model of a normal traffic light control strategy at intersection Ii.
 TABLE I. MEANINGS OF TRANSITIONS AND PLACES AND TRANSITION Now we present the two-level emergency traffic light
DELAYS OF FIG. 3
control strategy at intersection Ii by using TPNs. We first
P Meaning T Meaning  (s) define dotted double sided arrows that connect several
pχ1 R tχ1 Changing R to GS and GR 4
transitions with one place as shown in Fig. 5 where each arrow
pχ2 YL tχ7 and tχ8 Changing GS and GR to YS 15
and YR, respectively is attached with an integer x +. Dotted double sided arrows
pχ3 GL tχ9 and tχ10 Changing YS and YR to GL 2 with the same x represent the arcs that have the same direction,
pχ4 GS tχ3 Changing GL to YL 10 i.e., both from transitions to the place or from the place to
pχ5 GR tχ4 Changing YL to R 2
pχ6 YS others Immediate ones with no transitions. For example, in Fig. 5, if x1=x2 and one arc is from
pχ7 YR meaning but for a control t1 to p (respectively, from p to t1), then another is from t2 to p
purpose (respectively, from p to t2). The traffic light control strategy
pχ6-pχ9,
pi1-p i8 No meaning but for a control purpose from direction χ is drawn in Fig. 7. The following discussion is
M1 M2
about the control strategies extended from the one in Fig. 7.
M0 M3 M4
pζ1 pζ1
(tξ1),(tξ5,tξ6) (tξ7,tξ8) (tξ9,tξ10),(tξ2) pζ1 (tξ3) pζ1
pξ4 pξ6
pξ3 pξ2
A. Warning lights
pχ1 pξ5 pξ7

(tξ4)
We deploy two kinds of warning sign facilities named
(tζ4)
pξ1
warning lights at intersections for sending warning signals to
pξ1
pξ1 (tζ3) pξ1 (tζ9,tζ10),(tζ2) (tζ7,tζ8)
pζ4
(tζ1),(tζ5,tζ6) vehicles. In fact, they are installed besides traffic lights at
pζ6
pζ2 pζ3 pχ1
intersections. They can be divided into two sets, and each set
pζ7 pζ5
M9 M8 M5
has three lights, as shown in Fig. 6. The straight, left and right
M7 M6
ban signal lights are denoted by biχ,L, biχ,S, and biχ,R, giving a
ban signal to vehicles that intend to turn left, drive straight and
Fig. 4. The reachability graph of the model in Fig. 3. turn right at the upstream of Ii, respectively; the warning signal
lights are denoted by wiχ,L, wiχ,S, and wiχ,R, to warn vehicles that
t1 p t2 intend to turn left, drive straight and turn right at the upstream

{
of Ii, respectively. Thus, when some road links from Ii are
t1 p t2 if x1=x2 interrupted, the lights can ban or warn vehicles that attempt to
x1 x2
º enter the downstream conflict sections. For simplicity, biχ,S, biχ,L,
biχ,R, wiχ,S, wiχ,L, and wiχ,R are also used to denote places in the
following TPNs model. In the following discussions, the traffic
t1 p t2 light strategies that adopt warning lights are called emergency
ones.
Fig. 5. A dotted double sided arrow.

B. Level-I Strategy
(a) (b) (c)
Level-I emergency traffic light control strategy adopting ban
(d) (e) (f)
signal lights is designed via a TPN as shown in Fig. 8, where
p, p{pi1-pi8}, and (pa, tχ1) and (pa, tχ,LSR) are two dotted
double sided arrows that are used to describe real arcs in Fig. 3.
We define χ as the reverse direction of χ, i.e., if χ is direction n
(respectively, e, s, and w), χ represents s (respectively, w, n,
and e). Now we discuss four emergency strategies as follows:
Fig. 6. Three ban signal lights and three warning ones installed with the
traffic lights at intersection Ii: (a) biχ,L, (b) biχ,S, (c) biχ,R, (d) wiχ,L, (e) wiχ,S, and (f)
1) The straight traffic flow from direction χ needs to be
wiχ,R. forbidden. It is realized by keeping the traffic light GSχ off. In
the TPN model, a token is deposited in pχ,S. Then biχ,S obtains a
token meaning that the straight ban signal light is turning on.
III. A TWO-LEVEL STRATEGY There are two situations that the emergency strategy is realized:
Based on the normal traffic light control strategy proposed  pχ4 contains a token which means GSχ is on. Then the
in the previous section, we use TPNs to design traffic light immediate transition tχ,S2 is enabled and fires such that
control strategies at intersections to deal with incidents and a token is transferred from pχ4 to pχ6. Hence, GSχ is
prevent their induced traffic congestion. Such traffic light immediately turned off and YRχ is turned on.
control strategies are called emergency ones. They need some  pχ4 contains no token which means GSχ is off at the
additional lights, i.e., ban lights and warning one. The moment. tχ5 is prevented from firing because of the
proposed emergency strategy is divided into two levels: a ban inhibitor arc linked from pχ,S. When pχ8 gets a token,
signal strategy and a warning signal one. When it is executed, tχ,S1 is enabled and fires such that the token is
a ban signal strategy adopting ban lights can stop the traffic transferred from pχ8 to pχ10. Hence, GSχ remains off.
flow driving towards some directions and a warning signal Notice that we use tχ2 to realize the synchronization of lights
strategy adopting warning lights gives traffic flow a GSχ and GRχ. Under this strategy, a token stays in pχ10 until tχ10
recommendation of not driving to certain directions. The fires, i.e., the time duration of GRχ is reached.
cooperation among normal traffic lights and warning lights are
modeled and verified by TPNs.
 2) The right-turning traffic flow from direction χ should be signal lights are turned off and traffic lights return to their
stopped. It is realized by keeping the traffic light GRχ off. In normal strategies from the emergency ones. We will discuss
the TPN model, a token is deposited in pχ,R. Then biχ,R obtains a that in detail by reachability graph analysis.
token meaning that the right ban signal light is turning on.
There are two situations that the emergency strategy is realized: pα
pχ8 tχ5 pχ4 tχ7 pχ6 tχ9 pχ10
 pχ5 contains a token which means GRχ is on. Then the
immediate transition tχ,R2 is enabled and fires such that tχ1 tχ2
a token is transferred from pχ5 to pχ7. Hence, the traffic pχ9 tχ6 pχ5 tχ8 pχ7 tχ10 pχ11
light is immediately turned from GRχ to YRχ. tχ4 tχ3
 pχ5 contains no token which means GRχ is off at the pχ1 pχ2 pχ3
moment. tχ6 is not enabled because of the inhibitor arc
linked from pχ,R. When pχ9 gets a token, tχ,R1 is enabled pβ
and fires such that the token is transferred from pχ7 to
Fig. 7. The TPN model of a strategy for direction χ extracted from Fig. 3.
pχ11. Hence, GRχ remains off.
3) The left-turning traffic flow from direction χ should be tχ,S1 tχ,L4
stopped. It is realized by keeping the traffic light GLχ off. In the
tχ,LSR pα tχ,S3 biχ,S tχ,S4 biχ,L
TPN model, a token is deposited in pχ,L. Then biχ,L obtains a
1 pχ,S tχ,S2 tχ,L3
token meaning that the left ban signal light is turning on. There pχ,LSR 1 pχ8 tχ5 pχ4 tχ7 pχ6 tχ9 pχ10 pχ,L
are two situations that the emergency strategy is realized:
tχ1 tχ2
 pχ3 contains a token which means GLχ is on. Then the pχ9 tχ6 pχ5 tχ8 pχ7 tχ10 pχ11
tχ4 tχ3
immediate transition tχ,L2 is enabled and fires such that
pχ1 pχ2 pχ3
a token is transferred from pχ3 to pχ2. Hence, GLχ is
immediately turned off and YLχ is turned on. In this 2
tχ,R3 biχ,R tχ,R4 tχ,L1
situation, both tχ2 and tχ3 do not fire. pχ,R tχ,R2

 pχ3 contains no token which means GLχ is off at the 2 pβ 2 tχ,R1

moment. tχ2 is not enabled because of the inhibitor arc 2 tχ,L2
linked from pχ,L. When both pχ10 and pχ11 get their
tokens, tχ,L1 is enabled and fires such that the token is Fig. 8. The TPN model of strategy I.
transferred from pχ10 and pχ11 to pχ1. Hence, GLχ
remains off.
In the operation of the traffic-light control model in Fig. 3,
during a traffic light transition cycle, a token is deposited into pχ,S' wiχ,S
or removed from p at direction χ. When GLχ is changed to YLχ, pχ8 tχ5 pχ4 tχ7 pχ6 tχ9 pχ10
i.e., tχ3 fires and a token is filled into or removed from p. In the
emergency one, however, tχ3 will not fire. Thus dotted double tχ1 tχ2
sided arrows (p, tχ,L1) and (p, tχ,L2) are drawn in order to pχ9 tχ6 pχ5 tχ8 pχ7 tχ10 pχ11
preserve the token count in p. wiχ,R pχ,R'

4) If there exists a token in pχ,LSR, all traffic flows from

pχ,L' wiχ,L
direction χ should be forbidden. It is realized by keeping all
traffic lights GSχ, GRχ, and GLχ off. In the TPN model, a token tχ4 tχ3
is deposited in pχ,LSR. At this situation, each of pχ,R, pχ,S, and pχ1 pχ2 pχ3
pχ,L contains a token. Then each of biχ,R, biχ,S and biχ,L gets a
token meaning that all ban signal lights are turning on. There Fig. 9. The TPN model of strategy II.
are two situations that the emergency strategy is realized:
pn,S pn,R pn,L pn,LSR pn,S' pn,R' pn,L'
 pχ1 contains a token which means Rχ is on, and tχ1 is
prevented from firing because of the inhibitor arc from
pχ,LSR. At this situation, tχ,LSR fires and a token is
pw,L' pe,S
deposited into or removed from p and p. Note that
pw,R' pe,R
both p and p are marked or neither is. Hence, Rχ is
pw,S' pe,L
kept on. tid tic pe,LSR
 pχ1 contains no token, due to the tokens in pχ,S, pχ,R and pw,LSR
pe,S'
pχ,L such that the traffic lights can be turned from green pw,L

ones to yellow ones, and then to red ones. Thus Rχ is pw,R Strategies pe,R'

turned on and will be kept on. pw,S pe,L'

Note that, in the strategy, both tχ1 and tχ3 will not fire. The
dotted double sided arrows are drawn to preserve the token
count in p and p.
ps,L' ps,R' ps,S' ps,LSR ps,L ps,R ps,S
When tokens in pχ,R pχ,S, and pχ,L are removed, biχ,R, biχ,S and
i Fig. 10. The dynamic TPN model of a traffic light strategy decision.
b χ,L will not contain token either. At this situation all ban
C. Level-II Strategy north turning left have a conflict. Also at M04 we have a
conflict. To solve this problem, we need to ensure the proper
Level-II emergency traffic light control strategy adopting

synchronization of tχ2 and tχ2 and we construct a model in Fig.
warning signal lights is designed via a TPN as shown in Fig. 9.
15. As a result, the reachability graph is constructed as shown
If there respectively exists a token in pχ,S', pχ,R', and pχ,L', the
in Fig. 16.
traffic flow from direction χ should obtain a warning for
vehicles not to run straightly, turn right, and turn left, Suppose that the traffic flows from all directions should be
respectively. When there is no token in pχ3, pχ4, or pχ5, a token forbidden, i.e., the state of the signal should be M0=(Rn, Re, Rs,
will be generated in wiχ,L, wiχ,S, or wiχ,R, respectively, which Rw). We should adopt the strategy in Fig. 17. In the TPN
means that left, straight, or right warning signal light is turned model as shown in Fig. 8, a transition firing sequence
on when the left, straight, or right traffic light is off. At this =(tξ,LSR),(tζ,LSR), can fire repeatedly where M0[>M0 as shown
situation, vehicles could change their directions by following in Fig. 18. It is a livelock and has no benefit to traffic control.
the warning signals. Note that more than one but not all of pχ,S', A recovery model is built in Fig. 19. If all pχ1 and pχ,LSR contain
pχ,R' and pχ,L' can contain a token. tokens, transition tML fires and a token is generated in pML.
Transitions tn,LSR, ts,LSR, te,LSR, and tw,LSR cannot fire because of
D. Strategies to prevent traffic congestion the inhibitor arcs from pML. Thus  is prevented. When the
emergency strategy stops, tokens in pχ,LSR are removed. Then
Now a TPN model of traffic light strategy is designed in Fig. one of tML1-tML4 fires by removing the token in pML, and the
10. Given the traffic flow condition, strategies (when and traffic lights return to their normal strategy.
where to operate the ban lights and warning ones) should be
obtained. Here we adopt certain strategies to decide the arcs to Notice that other strategies can be obtained and verified in a
similar way when the straight, left, or right traffic flow from
be added to connect pχ,S, pχ,R, pχ,L, pχ,LSR, pχ,S', pχ,R', and pχ,L'
one of the four directions needs to be forbidden while others go
with transitions tid and tic. If a token needs to be deposited in
under their normal way. Thus the correctness of the TPN
p{pχ,S, pχ,R, pχ,L, pχ,LSR}, two arcs (tid, p) and (tic, p) are drawn.
model in Fig. 8 is verified. By using the reachability graph
Note that the detailed emergency strategies are determined by
analysis method we can verify the correctness of the TPN
the position of the incident which will be presented in the
model in Fig. 9 as well.
following section.

tid pn,S tic

E. Property Analysis
When a TPN model is built, We can verify the correctness of
Fig. 11. A dynamic TPN model of traffic light strategy decision.
the TPN-based traffic light strategy by using a reachability
graph analysis method. Here a reachability graph is pζ1
M20 M23
constructed with each arc corresponding to a sequence of tic,tn,S4
ps4 (ts7,tξ8) M25
bin,S tic,tn,S4 tn9
transitions containing only one timed one. Besides, these (tξ1),(ts5,tξ6,tn,S1)
pζ1
M23 pζ1 pξ5
pζ1
transitions that fire at the same time are put in parentheses. M11
tn9
ps4 (ts7,tξ8) M21
pn6 (ts7,tξ8,tn9)
ps4
bin,S M3
This way allows us to reduce the size of the graph greatly. The pζ1
pξ5 bin,S
ps4
(ts7,tξ8) pζ1
tn9 pξ5
(ts7,tξ8,tn9) pξ5 (tξ9,tξ10),(tξ2)
reachability analysis can also be done through a tool such as pn6 M12
pζ1
M22
pξ4
ps4 pξ5 M24
TimeNet [44]. ps4 (ts7,tξ8) bin,S tn9
pξ5
(ts9,tξ10),(tξ2) M13 M14
M10 pξ5 pζ1 bin,S bin,S
tic,tn,S4
We now give an emergency strategy as an example. Suppose bin,S pξ4 pζ1 (tξ3) pζ1
tic,tn,S4 M25
that the straight traffic flow from the North direction needs to pχ1
tid,tn,S3,tn,S2 pξ5 pξ3 pξ2 (tξ4)
M24
be forbidden while others go under their normal way. The TPN tic,tn,S4 tid,tn,S3 tid,tn,S3 tic,tn,S4 tid,tn,S3 tic,tn,S4 tid,tn,S3
model of such strategy generated from the model in Fig. 10 is M0 M1 M2 M3 M4
built as shown in Fig. 11, where there is only one input place of
tic and output place of tid, i.e., pn,S. Under such situation, we can M9 M8 M7 M6 M5
obtain the reachability graph of the model in Fig. 8 as shown in
tic,tn,S4 tid,tn,S3 tic,tn,S4 tid,tn,S3 tic,tn,S4 tid,tn,S3 tic,tn,S4 tid,tn,S3 tic,tn,S4 tid,tn,S3
Fig. 12. Since this graph is a finite circle containing 26
markings M0-M25 corresponding to 26 traffic light states that (tζ4)
bin,S
(tζ3)
bin,S
(tζ9,tζ10),(tζ2)
bin,S
(tζ7,tζ8)
bin,S
(tζ1),(tζ5,tζ6)
bin,S

are to ensure the traffic flow control, the TPN model is pξ1 pξ1 pξ1
pζ6
pξ1
pζ4
pχ1

reversible. Similarly, we can verify the correctness of the other pζ2 pζ3
pζ7 pζ5
M15
M19 M18
strategies by deriving their reachability graph. M17 M16

By using reachability graph analysis, we find some

Fig. 12. The reachability graph of the emergency strategy of Fig. 8 under the
situations where conflicting traffic flows appear. There are two condition of Fig. 11.
situations: both the straight and right traffic flows from
direction χ need to be forbidden as shown in Fig. 13 (a), and pχ,R
pχ,R
the straight and right traffic flows from direction χ and the right tid tic
tid tic
(or left) traffic flow from direction χ need to be forbidden while pχ,R(pχ,L)
others go under their normal way as shown in Fig. 13 (b). We pχ,S pχ,S
take Fig. 13 (b) as an example and construct its reachability
graph. In the graph there is a state M00 as shown in Fig. 14
(a) (b)
where GSs and GLn are on at the same time. Permitted by such
strategy, traffic flows from south driving straight and from Fig. 13. Two dynamic TPN models of traffic light strategy decision.
 M02
M00 M03 M04 tML
bis,R
(tn9,tξ10),tn2
bis,R bin,R bis,R bis,R
M01
bin,R bin,S bin,R bin,R tn,LSR te,LSR
bin,S bis,R pζ1 ts7 bin,S (tξ10,ts11),tn2 bin,S
pζ1 (tξ1),(tn,S1,tn,R1,ts,R1),tn2 bin,R ps4 pζ1 pζ1

ps4 bin,S pn6 pξ6 ps6 pML

pn3 pχ1 pξ7 pξ7 pn3 ps1 pe1 pn1 pw1 ps,LSR pe,LSR pn,LSR pw,LSR
tw,LSR ts,LSR
tid,(tn,S3,tn,R3,ts,R3) tid,(tn,S3,tn,R3,ts,R3),(tn,S2,tn,R2,ts,R2) tML1
tML2
tML3 t
M0 M1 ML4

Fig. 14. The reachablity graph under strategy Fig. 13 (b).

Fig. 19. A livelock prevention model.

pχ4 pχ6 IV. CASE STUDY

We have designed emergency traffic light strategies by
tχ2 using a TPN model and verified that such strategies can be
correctly implemented with co-design of hardware and
software. Now we evaluate the effectiveness of the proposed
pχ5 pχ7 strategy by employing MATLAB software through simulation..
First we build a two-way rectangular grid networks based on a
cell transmission model (CTM) [45], [46]. Then we extend the
Fig. 15. A synchronization model. model in [47] by considering the vehicles route-changing rate
and traffic signal strategy. Through the simulation of our
M02
M03 M04
designed model, we identify the influences of some important
M00 bis,R
(tn9,tξ10) parameters, i.e., route-changing behaviors of drivers, the time
bis,R bin,R bis,R bis,R
M01 to operate the emergency strategy, and the inflow of the traffic
i
b n,S i
bin,R
bin,R b n,R
network.
bin,S bis,R pζ1 ts7 bin,S (tξ10,ts11) bin,S
pζ1 (tξ1),(tn,S1,tn,R1,ts,R1) bin,R ps4 pζ1 pζ1

ps4 bin,S pn6 pξ6 ps6 A. Cell Transmission Model

pχ1 pξ7 pξ7
Daganzo [45] [46] proposes CTM to simplify the solution
scheme of the Lighthill-Whitham-Richards (LWR) model [48]
tid,(tn,S3,tn,R3,ts,R3) tid,(tn,S3,tn,R3,ts,R3),(tn,S2,tn,R2,ts,R2)
[49] by adopting the following relationship between traffic
M0 M1 flow q (vehicles per hour) and density k (vehicles per
lane-mile) as shown in Fig. 20:
Fig. 16. A part of reachability graph of the emergency strategy of Fig. 8
under the condition of Fig. 13 (b).
q=min{vk, qmax, w(kJ-k)}, 0≤k≤kJ
where v (miles per hour) is the free flow speed and w (miles
per hour) is the speed of all backward moving waves, and qmax
pe,LSR and kJ denote the inflow capacity (or maximum allowable
pn,LSR inflow) and jam density, respectively. We have the following
tid tic CTM:
yi(t)=min{ni-1(t, Qi(t), w(Ni(t)-ni(t))/v}
pw,LSR
ni(t+1)=ni(t)+yi(t)-yi+1(t)
ps,LSR
FLOW, q

Fig. 17. A dynamic TPN model of traffic light strategy decision.

M qmax
(tξ,LSR),(tζ,LSR)
i
b χ,R v -w
tid,(tχ,S3,tχ,R3,tχ,L3)
biχ,L
M0
biχ,S
tic,(tχ,S4,tχ,R4,tχ,L4)
pχ1
O DENSITY, k kJ

Fig. 18. A part of reachability graph of the emergency strategy of Fig. 8

under the condition of Fig. 17. Fig. 20. Flow-density relationship for the generalized CTM [22].
where yi(t) is the number of vehicles that flow into cell i
during time interval t, ni(t) is the number of vehicles in cell i Upstream reservoir Downstream
queue area
before t, Ni(t) denotes the maximum number of vehicles that
qL
can be contained in cell i during t, and Qi(t) denotes the inflow ... qS
capacity in cell i during t.
qR

1 ... λ-1 λ
B. The traffic network and its CTM
Fig. 19. Link a in two-way grid networks.
In the networks as shown in Fig. 19, each link a is divided
into two distinct zones: downstream queue storage areas qL, qS
and qR where vehicles are organized into separate turning N
movements (left in qL, straight in qS, and right in qR), and an
upstream reservoir where the turning movements are mixed.
We assume that the upstream reservoir is composed of λ cells,
and the length of the channelized queue area is short and
equals the length of a cell (λ+1). Note that the length of the
channelized downstream queue area is set to one cell because
B
of the results in [47] that the congestion time is reduced when
the channelized downstream queue area is very short. As
shown in Fig. 21, we construct an 8×8 two-way grid network C A
Incident
to test the model. All boundary nodes are both origins and
destinations. L, S, and R denote, respectively, the D
proportions of vehicles travelling in the left turning direction,
in the ahead direction and in the right turning direction. Stop
line width assignment variables αL, αS, αR denote, respectively,
the proportions of the segregated queue areas devoted to the
left turning queue storage area, to the ahead queue storage area
and to the right turning queue storage area. According to the
definition, we have L+S+R=1 and αL+αS+αR=1. We adopt a
‘balanced’ layout of stop line assignment [1] such that the stop
line widths devoted to the ahead direction and to the turning
direction are in exactly the same ratio as the demands, i.e.,
L=αL, S=αS, and R=αR. Fig. 21. An 8×8 two-way grid network.

Table II describes the symbols used in the following content. TABLE II. SYMBOLS AND THEIR MEANINGS
Note that traffic rules will not be changed and the vehicles in
Symbols Meaning
the downstream queue areas will not change lanes. Using the
nia(t) The number of vehicles contained in cell i of link a at the start
network model, we design a network traffic simulation model of time interval t
based on the time-step method. The inflow formulation can be niab(t) The number of vehicles contained in cell i of link a and take
classified into three categories: inflow of upstream link b as the next link at the start of time interval t
reservoir(i=1), inflow of upstream cells (1< i ≤λ-1), and inflow yia(t) The number of vehicles that flow into cell i of link a in time
interval t
of channelized downstream queue area (i=λ). We consider the yiab(t) The number of vehicles that flow into cell i of link a and take
influence of traffic flow lane changing behavior to illustrate link b as the next link in time interval t
the designed traffic light strategy. The inflow formulation is Qia(t) The maximum number of vehicles that can flow into cell i of
presented as follows: link a in time interval t
Nia(t) The maximum number of vehicles that can be present in cell i of
1) Inflow of upstream cells link a in time interval t
ab Proportion of vehicles traveling from link a to link b
The inflow of upstream cells can be calculated by: αab Proportion of stopline width devoted to vehicles traveling from
yia(t)=min{ni-1a(t), Q(t), w(N-nia(t))/v}, 1<i≤λ-1 (1) link a to link b
yiab(t)=abyia(t), 1<i≤λ-1 (2) dia(t) The sum of the congestion delay of vehicles contained in cell i
of link a at time interval t
2) Inflow of channelized downstream queue area d(t) The total congestion delay of vehicles contained in the whole
network at time interval t
The up bound of inflow of the downstream queues area for
vehicles travelling from link a to link b is computed as
follows: The inflow of each direction can be calculated as follows

y'ab(t)=min{αabQ(t), w(αabN-nλab(t))/v} (3) yλab(t)=min{abyλa(t), αabnλ-1a(t)} (5)

Because of interference between turning vehicles and ahead We model the proposed traffic light strategy in the CTM
vehicles [1], the total inflow of the channelized queues area model. We suppose that in our proposed two-level strategy, in
can be formulated as follows each direction, there is at most one warning light or ban light
being on. We suppose that when seeing a ban signal at the
yλa(t)=minbB {y'ab(t)/αab}
m (4) direction where vehicles intend to drive, d1 percent of them
change their routes, while seeing a warning signal, d2 percent
do so. Thus we have the proportion of traffic flow in the in Fig. 22 which demonstrates Property 1, i.e., if Γ<δQ0.833,
segregated queue area computed as follows. When a warning e.g. Γ=0.80 and 0.83, there is no congestion.
light at direction b is on and the other two directions are
We install a single incident in our simulation: a single
supposed to be c and d. Let 'ab denote the proportion of
incident occurs on the 5th cell of a link in the network as
vehicles traveling from link a to link b. At a ban or warning
shown in Fig. 21. We construct emergency strategies: ban
light-on situation, we have
signal strategy works at intersection A while warning signal
'ab=ab(1-di), 'ac=ac+abdi/2, and 'ad=ad+abdi/2 (6) strategies work at intersections B, C and D. In detail, a
dynamic TPN model in Fig. 10 can be constructed to execute
where i=1 if the light is a ban light and i=2 if it is a warning the strategy: there are arcs from tAd to pn,L, pw,S, and ps,R (also
light. denoted by pAn,L, pAw,S, and pAs,R), and from these places to tAc.
If such lights are on at direction b', the total inflow of the Tokens are filled in pAn,L, pAw,S, and pAs,R. Similarly, arcs are
channelized queues area can be obtained via (3)-(5), where added from tBd to pBe,L', pBn,S', and pBw,R' and from these places
ab='ab is computed by (6). to tBc; from tCd to pCn,L', pCw,S', and pCs,R' and from these places
to tCc; and from tDd to pDw,L', pDs,S', and pDe,R' and from these
3) Inflow of upstream reservoir and outflow of channelized places to tDc.
downstream queue area.
TABLE III. PARAMETERS FOR THE CTM [47]
yλ+1 is defined as the outflow of the terminal cells.
Parameters Values
yλ+1ab(t)=min{nλab(t), αabQλ(t), w(N-nb1(t))/v} (7) The length of each time interval 5 s
Jam density 133 vehicles/km/lane (i.e., 7.5 m for every
y1a(t)=bA yλ+1ba(t)
l (8) vehicle in each lane)
Free-flow speed 54 km/h (i.e., 15 m/s), and backward
yλ+1a(t)=bBmyλ+1ab(t) (9) shock-wave speed: 21.6 km/h (i.e., 6 m/s)
The update of the number of vehicles contained in each cell, Number of lanes 2
is formulated as follows Flow capacity 1800 vehicles/h/lane (i.e., 2.5
vehicles/time interval/cell)
nia(t+1)=nia(t)+yia(t)-yi+1a(t), 1≤i≤λ (10) Cell length 75 m, and the holding capacity of each
Traffic signal control and incidents are modeled by cell is 20 vehicles.
The number of cells of each 9 (i.e., the length of every link is 675m)
modifying the value of the corresponding flow capacity of the link
affected cells. Qλab(t)=0 if t belongs to the period when red
traffic light is on. At this time, vehicles running to link b in
cell λ of link a are forbidden. Qia(t)=0 if t belongs to the period
with an obstruction on cell i. We assume that Qia(t) and Nia(t)
are independent of cell number and time, i.e., Qia(t)=Q and
Nia(t)=N.
Traffic jam size is used to describe the effect of congestion.
A cell is called a jammed one if the density of it in the
upstream reservoir or in any direction of the downstream
channelized area is greater than 0.9N [47]. The size of traffic
jams is described in terms of the total number of jammed cells.
Fig. 22. Congestion formulation when Γ=0.80, 0.83, 0.84, 0.86, and 0.90.
The analysis period of interest is divided into 2000 intervals
(i.e., 2.78 h). The traffic demand is Γ vehicles per interval for
each origin. Define δ as the ratio that the green signal time C. Simulation results
divides the traffic light cycle time, we have a property as
First, we study the effect of the driving route changing rate
follows.
d1 and d2 on jam formation and dissipation. The incident
Property 1: Congestion appears if Γ>δQ. occurs at the 301st interval and is cleared at the 1000th interval
where the traffic demand is set to be Γ=0.5. Simulation results
Note that during a cycle T, the whole amount of traffic of congestion formulation and dissipation are shown in Figs.
riding in a lane and intending to driving left from the original 23-24. We can learn that under no strategy at this traffic
is Γ*T*αL while according to formula (7) the outflow from the situation, congestion begins at the 341th time interval; the
lane is less that δ*T*Q*αL. If Γ>δQ, the inflow overweight the maximum jammed cell count is 58 at the 1,000 th time interval;
outflow and the downstream queue storage area qL will finally and the congestion dissipation takes in total 1966 time
be blocked. The blockage of qL will lead to the congestion of intervals. Fig. 23 also shows the results under the proposed
the whole road. strategy where it begins from 301st interval to 1000th interval
The flow proportions for all directions are: L=0.2, S=0.5, in cases of d1=0.9, and d2=0.4, 0.5, 0.6, 0.7 and 0.9. The
and R=0.3. The initial network is empty, and some time maximum jammed cell count is no more than 10 if the turning
intervals are required to allow the system to stabilize. We rate d2 is not less than 0.5. At this situation, less that 350 time
suppose that the traffic inflow and outflow of the original intervals are needed to completely dissipate the congestion.
points are restricted by the same traffic light signal as well as Fig. 24 shows the results when d2=0.5 but d1 changes, i.e., 0.80,
the traffic light time duration as described in Fig. 3. We 0.85, 0.90, 0.95 and 1.00. We can learn that the maximum
evaluate the impact of factor Γ on traffic congestion as shown
jammed cell count is no more than 10 if d1 0.9. If d1  0.98,
there is no congestion.
Second, we study the effect of the start time to adopt the
emergency strategy. Under all emergency strategies, the time
needed for the dissipation is much less than the self-evolved
traffic. We obtain the result from Fig. 25 that under the
constant inflow Γ=0.5, the turning rates d10.9 and d20.5. If
the reaction time towards the incident is less than 10 minutes
(i.e., at the 421st time interval), the maximum jammed cell
count is no more than 10 and less than 350 time intervals are
needed to completely dissipate the congestion. Thus the
Fig. 25. Congestion formulation and dissipation under no strategy and
emergency strategy indeed offers excellent performance. emergency ones such that the strategy begins from 301 st, 337th, 361st, 421st,
and 481st intervals, respectively, to 1000th interval and d1=0.9, and d2=0.5.
Third, we study the effect of the time to stop the emergency
strategy that begins from the 421st interval. From Fig. 26 we
can see that the congestion starts to dissipate at the stop of the
emergency strategy. For example, if the end time of the
strategy is 10 minutes after the clearance of the incident (i.e. at
the 1121st time interval), the maximum jammed cell count is
12 and more than 600 time intervals are needed to completely
dissipate the congestion. Thus the emergency strategy should
be stopped as soon as possible when the incident is cleared
because the ban signal strategy causes congestion as well.
Finally, we evaluate the impact of factor Γ as shown in Fig.
27 when the strategy begins from the 301 st interval to 1000th
interval and d1=0.9, and d2=0.5. We can see that if Γ≤0.6, the Fig. 26. Congestion formulation and dissipation under no strategy and
emergency strategy can be accepted. If not, it causes emergency ones such that the strategy begins from 421 st interval to 1000th,
1037th, 1061st, 1121st, and 1181st intervals, respectively, and d1=0.9, and
congestion and takes more than 800 time intervals to dissipate. d2=0.5.
In conclusion, the emergency strategies can be adopted to
deliver highly desired performance under the following
conditions that the constant inflow Γ≤0.5, the turning rates
d10.9 and d20.5 and the incident reaction time is no more
than 10 minutes.

Fig. 27. Congestion formulation when Γ=0.4, 0.5, 0.6, 0.7, 0.8, and 1.0 and
the strategy begins from 301st interval to 1000th interval and d1=0.9, and
d2=0.5.

V. CONCLUSIONS
Fig. 23. Congestion formulation and dissipation under no strategy and
emergency ones such that the strategy begins from 301 st interval to 1000th This work designs a traffic-light control system at an urban
interval and d1=0.9, and d2=0.4, 0.5, 0.6, 0.7 and 0.9. traffic intersection to prevent large-scale traffic congestion
caused by incidents. Normal and emergency traffic light
control strategies are formulated. Their control logic is
presented in detail and proved to be valid by using timed Petri
nets. This work verifies the effectiveness of the strategy
through a simulation study where an extended CTM model is
used to construct the traffic network. The strategies perform
well in response to an incident under certain conditions
including timely termination of their use after the clearance of
the incident. The results can be used to improve the state of
the art in preventing urban road traffic congestion caused by
incidents. Future work intends to study the applications of the
proposed strategies in other complicated traffic networks.
Fig. 24. Congestion formulation and dissipation under no strategy and
emergency ones such that the strategy begins from 301 st interval to 1000th
interval and d1=0.80, 0.85, 0.90, 0.95 and 1.00 , and d2=0.5.
 REFERENCES no. 1, pp. 1159-1176, Nov. 2012.
[1] C. Wright and P. Roberg, “The conceptual structure of traffic jams,” [23] P. Roberg, “Distributed strategy for eliminating incident-based traffic
Transp. Policy, vol. 5, no. 1, pp. 23-35, 1998. jams from urban networks,” Traffic Eng. Control, vol. 36, no. 6, pp.
348-355, 1995.
[2] H. Al-Deek, T. W. P. Lochrane, C. V. Srinivasa, R. Chandra, and A.
Khattak, “Diversion during unexpected congestion on toll roads: the [24] J. Long, Z. Gao, P. Orenstein, and H. Ren, “Control strategies for
role of traffic information displayed on dynamic message signs,” IET dispersing incident-based traffic jams in two-way grid networks,” IEEE
Intell. Transp. Syst., vol. 6, no. 2, pp. 97-106, 2012. Trans. Intell. Transp. Syst. vol. 13, no. 2, pp. 469-481, Oct. 2012.
[3] V. Milanes, J. Godoy, J. Villagra, and J. Perez, “Automated on-ramp [25] M. C. Zhou and F. DiCesare, Petri Net Synthesis for Discrete Event
Control of Manufacturing Systems. Norwell, MA, USA: Kluwer, 1993.
merging system for congested traffic situations,” IEEE Trans. Intell.
Transp. Syst., vol. 12, no. 2, pp. 500-508, Dec. 2011. [26] M. C. Zhou and K. Venkatesh, Modeling, Simulation and Control of
[4] P. Zhang, C. X. Wu, and S. C. Wong, “A semi-discrete model and its Flexible Manufacturing Systems: A Petri Net Approach. Singapore,
World Scientific, 1998.
approach to a solution for a wide moving jam in traffic flow,” Physica
A, vol. 391, no. 3, pp. 456-463, 2012. [27] K. M. Ng, M. B. I. Reaz and M. A. M. Ali, “A review on the
applications of Petri nets in modeling, analysis and control of urban
[5] R. Bauza and J. Gozalvez, “Traffic congestion detection in large-scale
traffics,” IEEE Trans. Intell. Transp. Syst., vol. 14, no. 2, pp. 858-870,
scenarios using vehicle-to-vehicle communications,” J. Netw. Comput.
Mar. 2013.
Appl., vol. 36, no. 5, pp. 1295-1307, 2013.
[28] G. F. List and M. Cetin, “Modeling traffic signal control using Petri
[6] F. Knorr, D. Baselt, M. Schreckenberg, and M. Mauve, “Reducing
nets,” IEEE Trans. Intell. Transp. Syst., vol.5, no. 3, pp. 177-187, Sep.
Traffic Jams via VANETs,” IEEE Trans. Veh. Technol., vol. 61, no. 8,
2004.
pp. 3490-3498, July 2012.
[7] F. Terroso-Saenz, M. Valdes-VelaC, R. Toledo-Moreo, and A. F. [29] Y. S. Huang and T. H. Chung, “Modeling and analysis of urban traffic
Gomez-Skarmeta, “A Cooperative Approach to Traffic Congestion lights control systems using timed CP-nets,” J. Inf. Sci. Eng., vol. 24,
no. 3, pp. 875-890, 2008.
Detection With Complex Event Processing and VANET,” IEEE Trans.
Intell. Transp. Syst., vol. 13, no. 2, pp. 914-929, Feb. 2012. [30] Y. S. Huang and T. H. Chung, “Modeling and analysis urban traffic
[8] M. Papageorgiou, C. Diakaki, V. Dinopoulou, A. Kotsialos, and Y. B. light control systems,” J. Chin. Inst.Eng., vol. 32, no. 1, pp. 85-95,
2009.
Wang, “Review of Road Traffic Control Strategies,” Proc. IEEE, vol.
91, no. 12, pp. 2043-2067, Dec. 2003. [31] Y. S. Huang, Y. S. Weng, and M. C. Zhou, “Modular design of urban
[9] D. I. Robertson and R. D. Bretherton, “Optimizing networks of traffic traffic-light control systems based on synchronized timed Petri nets,”
signals in real-time - the SCOOT method,” IEEE Trans. Veh. Technol., IEEE Trans. Intell. Transp. Syst., vol. 15, no. 2, pp. 530-539, Nov.
vol. 40, pp. 11-15, Feb. 1991. 2013.
[10] P. R. Lowrie, “SCATS: A traffic responsive method for controlling [32] Y. S. Huang, Y. S. Weng and M. C. Zhou, “Critical scenarios and their
urban traffic,” Tech. rep., Roads and Traffic Authority, NSW, Australia, identification in parallel railroad level crossing traffic emergency
Traffic Control Section, P. O. Box 693, Darlinghurst, NSW 2010, systems,” IEEE Trans. Intell. Transp. Syst., vol. 11 , no. 4, pp. 968-977,
Australia, 1990. Oct. 2010.
[33] Y. S. Huang, Y. S. Weng, and M. C. Zhou, “Design of Traffic Safety
[11] M. Dotoli, M. P. Fanti, C. Meloni, “A signal timing plan formulation
Control Systems for Emergency Vehicle Preemption Using Timed Petri
for urban traffic control”, Control Eng. Pract., vol. 14, no. 11, pp.
1297-1311, Nov. 2006. Nets,” IEEE Trans. Intell. Transp. Syst., vol. 16, no. 4, pp. 2113-2120,
Mar. 2015.
[12] S. F. Cheng, M. A. Epelman, and R. L. Smith, “CoSIGN: A parallel
[34] L. Qi, W. J. Luan and M. C. Zhou, “Emergency Traffic-Light Control
algorithm for coordinated traffic signal control,” IEEE Trans. Intell.
Transp. Syst., vol. 7, no. 4, pp. 551-564, Dec. 2006. System Design for Intersections Subject to Accidents,” IEEE Trans.
Intell. Transp. Syst., vol. 17, no. 1, pp. 170-183, Sep. 2016.
[13] B. D. Schutter, and B. D. Moor. “Optimal traffic light control for a
single intersection,” Eur. J. Control, vol. 4, no. 3 pp. 260-276, 1998. [35] M. P. Fanti, A.M. Mangini, G. Iacobellis, W. Ukovich, “Freeway
Traffic Modeling and Control in a First Order Hybrid Petri Net
[14] K. Aboudolas, M. Papageorgiou, and E. Kosmatopoulos, “Store-and Framework,” IEEE Trans. Autom. Sci. Eng., vol. 11, no. 1, pp. 90-102,
forward based methods for the signal control problem in large-scale April 2014.
congested urban road networks,” Transp. Res. C, Emerg. Technol., vol.
17, no. 2, pp. 163-174, Apr. 2009. [36] J. Q. Wang, J. X. Yan, and L. X. Li, “Microscopic Modeling of a
Signalized Traffic Intersection Using Timed Petri Nets,” IEEE Trans.
[15] K. Aboudolas, M. Papageorgiou, A. Kouvelas, and E. Kosmatopoulos, Intell. Transp. Syst., vol. 17, no. 2, pp. 305-312, Feb. 2016.
“Store-and forward based methods for the signal control problem in
large-scale congested urban road networks,” Transp. Res. C, Emerg. [37] A. D. Febbraro, D. Giglio, and N. Sacco, “A Deterministic and
Technol., vol. 18, no. 5, pp. 680-694, Apr. 2010. Stochastic Petri Net Model for Traffic-Responsive Signaling Control in
Urban Areas,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 2, pp.
[16] A. Kouvelas, K. Aboudolas, M. Papageorgiou, and E. Kosmatopoulos, 510-524, Feb. 2016.
“A hybrid strategy for real-time traffic signal control of urban road
networks,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 3, pp. 884-894, [38] G. F. List and M. Mashayekhi, “A Modular Colored Stochastic Petri
Mar. 2011. Net for Modeling and Analysis of Signalized Intersections,” IEEE
Trans. Intell. Transp. Syst., vol. 17, no. 3, pp. 701-713, Mar. 2016.
[17] S. Zhao, Y. Chen, and J. A. Farrell, “High-Precision Vehicle Navigation
[39] L. Qi, W. J. Luan and M. C. Zhou, “Impact of Driving Behavior on
in Urban Environments Using an MEM’s IMU and Single-Frequency
Traffic Delay at a Congested Signalized Intersection,” IEEE Trans.
GPS Receiver,” IEEE Trans. Intell. Transp. Syst., vol. 17, no. 10, pp.
Intell. Transp. Syst. (published online)
2854-2867, April 2016.
[40] P. Roberg, C. R. Abbess, and C. Wright, “Traffic jam simulation,”
[18] N. Geroliminis, J. Haddad, and M. Ramezani. “Optimal perimeter
Journal of Maps, vol. 3, no. 1, pp. 107-121, 2007.
control for two urban regions with macroscopic fundamental diagrams:
A model predictive approach,” IEEE Trans. Intell. Transp. Syst., vol. 14, [41] G. Liu, “Complexity of the deadlock problem for Petri nets modeling
no. 1, pp. 348-359, Nov. 2013. resource allocation systems,” Inf. Sci., vol. 363, pp. 190–197, 2016.
[19] M. Ramezani, J. Haddad, and N. Geroliminis, “Dynamics of [42] G. J. Liu and C. J. Jiang, “Net-structure-based conditions to decide
heterogeneity in urban networks: aggregated traffic modeling and compatibility and weak compatibility for a class of inter-organizational
hierarchical control,” Transp. Res. B, Methodol., vol. 74, pp. 1-19, workflow nets,” Sci. China Inf. Sci., vol. 58, no. 7, pp. 1–16, 2015.
2015. [43] X. Lu, M. Zhou, A. C. Ammari, and J. Ji, “Hybrid Petri nets for
[20] J. Haddad, M. Ramezani, and N. Geroliminis, “Cooperative traffic modeling and analysis of microgrid systems,” IEEE/CAA J. Autom.
control of a mixed network with two urban regions and a freeway,” Sinica, vol. 3, no. 4, pp. 347–354, Oct. 2016.
Transp. Res. B, Methodol., vol. 54, no. 8, pp. 17-36, Aug. 2013. [44] A. Zimmermann, M. Knoke, A. Huck, and G. Hommel, “Towards
[21] Y. Ji and N. Geroliminis, “On the spatial partitioning of urban version 4.0 of TimeNET,” in Proc. 13th GI/ITG Conference Measuring,
transportation networks,” Transp. Res. B, Methodol., vol. 46, no. 10, pp. Modelling and Evaluation of Computer and Communication Systems,
1639-1656, Dec. 2012. 2006, pp. 477-480.
[22] J. Haddad and N. Geroliminis, “On the stability of traffic perimeter [45] C. F. Daganzo, “The cell transmission model: a simple dynamic
control in two-region urban cities,” Transp. Res. B, Methodol., vol. 46, representation of highway traffic,” Transp. Res. B, vol. 28, no. 4, pp.
 269-287, 1994.
[46] C. F. Daganzo, “The cell transmission model, part II: network traffic,”
Transp. Res. B, vol. 29, no. 2, pp. 79-93, 1995.
[47] J. C. Long, Z. Y. Gao, X. M. Zhao, A. P. Lian, and P. Orenstein,
“Urban Traffic Jam Simulation Based on the Cell Transmission Model,”
Netw. Spat. Econ., vol. 11, no. 1, pp. 43-64, 2011.
[48] M. Lighthill and G. Whitman, “On kinematic waves II: a traffic flow
theory on long crowded roads,” in Proc. Roy. Soc. London Series A, vol.
229, 1955, pp. 317-345.
[49] P. I. Richards, “Shock waves on the highway,” Oper. Res., vol. 4, no. 1,
pp. 42-51, Feb. 1956.

Liang Qi (S’16) received the B.S. and M.S. degrees

from Shandong University of Science and Technology,
Qingdao, China, in 2009 and 2012, respectively. He is
currently working toward the Ph.D. degree with the
Department of Computer Science and Technology,
Tongji University, Shanghai, China. He has been
supported by a scholarship from the China Scholarship
Council. From 2015 to 2016, he was a joint Ph.D.
Student with the Department of Electrical and
Computer Engineering, New Jersey Institute of
Technology, Newark, NJ, USA. He has authored over a dozen technical
papers in journals and conference proceedings. His research interests include
Petri nets, intelligent transportation systems, and traffic simulation.

MengChu Zhou (S’88-M’90-SM’93-F’03) received

the B.S. degree in control engineering from Nanjing
University of Science and Technology, Nanjing, China,
in 1983; the M.S. degree in automatic control from
Beijing Institute of Technology, Beijing, China, in
1986; and the Ph.D. degree in computer and systems
engineering from Rensselaer Polytechnic Institute,
Troy, NY, USA, in 1990. He joined New Jersey
Institute of Technology, Newark, NJ, USA, in 1990,
where he is currently a Distinguished Professor of
Electrical and Computer Engineering. He has authored over 680 publications,
including 12 books, over 360 journal papers (over 260 in IEEE Transactions),
and 28 book chapters. His research interests are in Petri nets, Internet of
Things, big data, Web services, manufacturing, transportation, and energy
systems. He is the Founding Editor of the IEEE Press Book Series Systems
Science and Engineering. He is a Life Member of Chinese Association for
Science and Technology-USA and served as its President in 1999. He is a
fellow of International Federation of Automatic Control and American
Association for the Advancement of Science. He is a recipient of the
Humboldt Research Award for US Senior Scientists, the Franklin V. Taylor
Memorial Award, and Norbert Wiener Award from IEEE Systems, Man, and
Cybernetics Society.

WenJing Luan (S’16) received the B.S. and M.S.

degrees from Shandong University of Science and
Technology, Qingdao, China, in 2009 and 2012,
respectively. She is now pursuing the Ph.D. degree in the
Department of Computer Science and Technology, Tongji
University, Shanghai, China. She received the Best
Student Paper Award-Finalist in the ICNSC 2016
conference. Her current research interests include Petri
nets, transportation systems and Smart City.