Professional Documents
Culture Documents
Tan 2017
Tan 2017
Abstract: This paper proposes a microscopic spatial-temporal method to capture the vehicle movement trajectories at an isolated inter-
section. A bi-objective programming model is presented to minimize the overall delay and emission increment with and without traffic signal
by jointly determining the optimal signal timing and speed limit. The proposed method explicitly incorporates the effects of vehicle speed
limit, acceleration, deceleration, and signal timing on the overall delay (including the delays caused by traffic signal, acceleration, and
deceleration) and emission increment with and without traffic signal control at the intersection. DOI: 10.1061/JTEPBS.0000101.
© 2017 American Society of Civil Engineers.
Author keywords: Signalized intersection; Overall delay; Traffic emission; Speed limit; Acceleration and deceleration.
k1
v1SW ¼ − v ð3Þ Evaluation of the Traffic Emission
km − k1
The evaluation of the traffic emission at an intersection is a chal-
and lenging issue because different emission estimations are associated
E ¼ E1 − E2 ¼ Fðuðx; tÞ; aÞdxdt equations of ðx 0 ; t 0 Þ. The coordinates of the point on C 0 0 D for the
SΔA 0 0 B 0 0 D vehicle
of ending to accelerate and passing through ðx 0 ; t 0 Þ are
ZZ 2 2
v2 v Z ffiffiffiffiffiffiffiffiffiffiffiffiffi
SW
þ xC þ ; t ∈ tD ; tC þ 1 1 v p
2aþ aþ × þ u 1 þ u2 du ð12Þ
a− aþ 0
00
for curve C D.
With those boundary curves, the emission in domain C 0 C 0 0 D Note that according to the definitions of speed of the shock-
can be analytically determined. Note that the instantaneous speed waves, v1SW and v2SW , given by Eqs. (3) and (4), respectively, the
is linearly increasing along any one parallel line of CD from a ratios v1SW =v and v2SW =v1SW are constant and related to the densities
certain level to v. Thus, the emission in domains C 0 CD and of the traffic flow. And thus, the total traffic emission E1 and E2 are
C 0 0 CD can be calculated via the curvilinear integral on the par- completely determined by the speed limit, acceleration rate, decel-
allel lines of CD and summarization along the curve C 0 C 0 0 . eration rate, and red time.
With h direct calibration, the pemission in i domain C 0 C 0 0 D is
ffiffiffiffiffiffiffiffiffiffiffiffiffi
−Þ þÞ
jCDj ∫ v0 ∫ u0 Fðw;a a− þ Fðw;a aþ 1 þ u2 dwdu , where jCDj ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxD − xC Þ2 þ ðtD − tC Þ2 . Bi-Objective Problem and Solution Properties
In summary, with the coordinates of C and D, given by Eqs. (1),
In this section, a bi-objective programming is proposed to mini-
(2), (5), and (6), respectively, the emission over the whole spatial-
mize the overall delay and emission increment, simultaneously,
temporal domain A 0 0 B 0 0 D, E1 , can be expressed as
by selection of the speed limit and signal timing. Namely, the
1 government sets the speed limit and signal timing, and the drivers
E1 ¼ ð−xC ÞrFð0; 0Þ select their deceleration and acceleration. It is assumed that all driv-
2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z pffiffiffiffiffiffiffiffiffiffiffiffiffi ers follow the same diving style and choose the same deceleration
1 þ ðv1SW Þ2 xC v Fðu; a− Þ 1 þ u2
þ du and acceleration. Once the deceleration and acceleration are given,
v1SW 0 a− the complete vehicle trajectories can be theoretically determined by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z pffiffiffiffiffiffiffiffiffiffiffiffiffi
1 þ ðv2SW Þ2 xC v Fðu; aþ Þ 1 þ u2 assuming that the initial speed is exactly equal to the speed limit, as
þ du discussed in the previous section, and the overall delay and emis-
v2SW 0 aþ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sion increment. The Pareto optimal solutions of the speed limit and
þ ðxD − xC Þ2 þ ðtD − tC Þ2 signal timing can also be determined by numerical method. How-
Z Z ever, it must be pointed out that the Pareto optimal solutions depend
v u Fðw; a Þ
− Fðw; aþ Þ pffiffiffiffiffiffiffiffiffiffiffiffiffi2 on the deceleration and acceleration adopted by the drivers, which
× þ 1 þ u dwdu
0 0 a− aþ are constant and exogenously given in the current paper. The
v1 bi-objective programming problem can be modelled as follows:
1 − SW
¼ v
r2 vFð0; 0Þ
2 1 − v1SW Dðv; g; rÞ
2 vSW min ð13Þ
v− ≤v≤vþ ;g≥0;r≥r− Eðv; g; rÞ
Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r v 1 þ ðv1SW Þ2 Fðu; a− Þ
þ v1 subject to the undersaturated condition of Eq. (8), where v− =
1 − vSW
2
0 a−
minimal speed to guarantee the normal flow rate of the intersection;
!
SW
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 þ ðv2SW Þ2 Fðu; aþ Þ pffiffiffiffiffiffiffiffiffiffiffiffiffi2 vþ = maximal speed to guarantee the safety at the intersection; and
þ 1 þ u du r− = minimal red time to allow the pedestrians safely crossing the
aþ intersection.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
The solution of Eq. (13), ðv ; g ; r Þ, is called the Pareto-
1 1 2 1 v1 1 2
v
2 a− þ aþ ðvSW Þ2 þ a1þ þ vSW 2 a efficient solution if there does not exist another feasible solution
SW −
þ v1SW
ðv; g; rÞ such that Dðv; g; rÞ ≤ Dðv ; g ; r Þ and Eðv; g; rÞ ≤
1 − v2 Eðv ; g ; r Þ with at least one strict inequality. Two methods can
Z Z
SW
Fðw; aþ Þ pffiffiffiffiffiffiffiffiffiffiffiffiffi2
v u Fðw; a Þ
be adopted to obtain the Pareto solution set for the Eq. (13) as also
−
× þ 1 þ u dwdu ð11Þ used by Guo and Yang (2009) and Tan et al. (2010). The simplest
0 0 a− aþ method is weighted-sum method as follows:
In Eq. (11), the first term is the emission in the queue, the second min λ1 Dðv; g; rÞ þ λ2 Eðv; g; rÞ
and third terms are the emission during the deceleration and accel- v− ≤v≤vþ ;g≥0;r≥r−
eration process with and without complete stop, respectively. In
particular, when the vehicles traverse the domain at the maximal The different Pareto optimal solution can be obtained by set-
cruise speed v during the green time, the total emission E2 is the ting different non-negative weight parameters λ1 and λ2 . In fact, the
product of the area of A 0 0 B 0 0 D and the emission rate Fðv; 0Þ or weighted objective is commonly used to investigate the trade-off of
calculated by Eq. (11) by setting Fð·; ·Þ at Fðv; 0Þ, i.e. the travel delay and emission (Ge et al. 2014).
Finally, the Pareto optimal speed limit v, green and red times, g Numerical Study
and r, can be jointly determined by considering the following
constraint programming In this section, a numerical example is adopted to depict the es-
sential idea of the paper. Without loss of generality, suppose the
min Eðv; g; rÞ
v− ≤v≤vþ ;g≥0;r≥r− density of arrival flow k1 ¼ 13 km and the density of the departure
flow k2 ¼ 12 km, where km is the jam density at the queue. With sim-
subject to
ple calibration, it is known that both the speeds of the shockwave
Dðv; g; rÞ ≤ D̄ after and before the queue, given by Eqs. (3) and (4), are v1SW ¼
− 12 v and v2SW ¼ −v, respectively. Without necessarily representing
or the following constraint programming a realistic setting, the deceleration and acceleration are assumed
min Dðv; g; rÞ to a− ¼ 2 m=s2 and aþ ¼ 3 m=s2 . According to Eqs. (7) and
v− ≤v≤vþ ;g≥0;r≥r− (8), the overall delay in the isolated intersection is D ¼
v 12 r þ 24
5
v 2 with the undersaturated condition g ≥ r þ 15 24 v.
subject to
To examine the spatial-temporal distribution of the emission
Eðv; g; rÞ ≤ Ē rate, the instantaneous speed of each vehicle passing through the
domain is first be derived with any given point. It must be pointed
When the value D̄ varies in ½Dmin ; Dmax or Ē in ½D̄min ; D̄max , out that the explicit formulation of the instantaneous speed is not
the solutions of those programming problems will figure out the necessarily required to calculate the emission increment according
Pareto solution set of original Eq. (13). If the time delay and vehicle to Eqs. (11) and (12). Note that the spatial-temporal trajectory with
emission are positive correlative to each other, the constraint pro- any given point on the boundary is completely determined. Given
gramming problems become unconstraint programming problems, the deceleration and acceleration, the spatial-temporal instantane-
and the optimal speed limit, green, and red times can minimize the ous speed of the vehicles can be explicitly determined. For the
delay and emission simultaneously. Otherwise, the trade-off Pareto vehicles with complete stop, the instantaneous speed is
optimal solution such that the constraint condition is binding and qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the objective is minimized under the constraint. uðx; tÞ ¼ v1SW þ ðv1SW Þ2 þ 2a− v1SW t − 2a− x ð14Þ
Note that the overall delay and the emission increment are free
from the effect of the selection of the green time provided that the
when ðx; tÞ satisfies
condition in Eq. (8) holds. Therefore, it is convenient to assume that
Eq. (8) is binding and the green time can be completely determined
v v2 a ðt − tC Þ2
by the red time and speed limit with given acceleration and decel- max v1SW t þ − ; xC − −
eration rates. That is to say, the bi-objective function can be reduced a− 2a− 2
2
to minimize both overall delay and emission increment by selection a t
≤ x ≤ min v1SW t; − − ;
of the red time and speed limit. 2
It is clear to see that the overall delay D, given by Eq. (7), is qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
convex and increasing with respect to v and r, because both the uðx; tÞ ¼ v2SW þ ðv2SW Þ2 þ 2aþ v2SW ðr − tÞ þ 2aþ x ð15Þ
ratios v1SW =v and v2SW =v1SW are constant. Furthermore, the overall
delay D is also convex and decreasing with respect to a− and aþ . when ðx; tÞ satisfies
50 14
0 12
Distance from the intersection (meter)
-50 10
-100 8
-150 6
-200 4
-250 2
-300 0
-10 0 10 20 30 40 50
Time (second)
are increasing with respect to the speed limit v and red time r. Thus, Fig. 7 shows the overall delay and emission increment with dif-
given the feasible domain v− ≤ v ≤ vþ and r ≥ r− , it is natural for ferent levels of the speed limit v and assuming the minimal red time
the planner to select the lowest speed limit v− and red time r− to r− ¼ 20 s for the safety of the pedestrians. It is clear to see that the
achieve the optimal performance or lowest levels of both overall overall delay is strictly increasing in v, whereas the emission incre-
delay and total emission at a signalized intersection. However, for ment increases by a little bit and then decreases. There is a trade-off
any given red time, the emission increment E first increases by a when design the speed limit for the isolated intersection. Fig. 8
little bit and then decreases significantly in the speed limit v, depicts the Pareto optimal frontier of bi-objective programming
whereas the overall delay D is strictly increasing in v. Thus, the Eq. (13) with red time r ¼ 20 s. In this case, the Pareto optimal
objective functions in programming Eq. (13) conflict with each frontier can be readily obtained by selecting the speed limit v.
other. Furthermore, the overall delay D is strictly increasing in In this numerical example, the Pareto frontier when given the
the red time r with given v, whereas the emission increment E red time is concave and changes from ð0; 0Þ to approximately
is strictly decreasing in the red time r with given v. For the bi- ð12; 380 veh × sec; −2; 565 gÞ in the ðD; EÞ-domain with the
objective programming Eq. (13) with constraints v− ≤ v ≤ vþ speed limit varies from 0 to 100 km=h. For the more practical fea-
and r ≥ r− , it is known that, at any Pareto-efficient solution, the red sible domain of the speed limit, such as 40–60 km=h, the objective
time r and the speed limit v are determined by the trade-off of D values of the delay and emission can be limited in the feasible do-
and E. main 40–60 km=h, and the corresponding Pareto frontier is shown
0 5
0 8
4
-100 -100 6
3
-200 -200 4
2
-300 -300 2
1
-400 -400 0
-10 0 10 20 30 40 50 60 -10 0 10 20 30 40 50 60
Downloaded from ascelibrary.org by Nanyang Technological University- Library on 10/12/17. Copyright ASCE. For personal use only; all rights reserved.
100 14
12 15
0 0
10
-100 8 -100
10
-200 6 -200
4 5
-300 -300
2
-400 0 -400
-10 0 10 20 30 40 50 60 -10 0 10 20 30 40 50 60
(c) Time (second) (a-=a+=3m/s 2) (d) Time (second) (a-=a+=4m/s ) 2
100
25
00
00
90
80 15
00
00
70
Red time r (second)
60 80
00
0
2000
50
40
00
0
40
30
10 0
10 0
20 00
50
20
50
10 0
00
00
00
10
0
1
0.1
10
0.0
20 0
1
0 5 20 00
0 10 20 30 40 50 60 70 80 90 100
Speed limit v (km/hr)
Fig. 5. Overall delay versus total emission with various levels of speed and red time
by the solid curve in Fig. 8, which is a partition of the whole Pareto cruise speed, deceleration and acceleration, and the signal timing
frontier. at that intersection. This paper developed a microscopic spatial-
temporal method to explicitly incorporate the effects of the ve-
hicles’ instantaneous speed (cruise speed, deceleration rate, and
Conclusions and Further Studies acceleration rate) and signal timing at an isolated intersection.
Based on this method, the analytical expressions are proposed for
It has been widely recognized that both the overall delay and traffic the overall delay and emission increment with and without traffic
emission at an intersection are significantly affected by the vehicle signal control at the intersection. The former captures the time
-1 0000
0.5
-2 000
90
-5 00
-1 00
-5
-0.1
80
70
28 80
50
Downloaded from ascelibrary.org by Nanyang Technological University- Library on 10/12/17. Copyright ASCE. For personal use only; all rights reserved.
40 53 8
0
0.1 0
30
20
-2 0
-5
10
-1
00
00
00
-5
-0 . 13 80 53 80
167 32 0 28 80
0 0
0 10 20 30 40 50 60 70 80 90 100
Speed limit v (km/hr)
Fig. 6. Overall delay versus emission increment with various levels of speed and red time
14000 500
Emission difference
12000 0
Overall delay D (vehicle second)
10000 -500
6000 -1500
4000 -2000
2000 -2500
0 -3000
0 10 20 30 40 50 60 70 80 90 100
Speed limit v (km/hr)
Fig. 7. Overall delay and emission increment with different speed limits
delay and the latter represents the traffic emission caused by the time, and first increasing by a little bit and then decreasing signifi-
signal. A bi-objective programming model is presented to jointly cantly in the cruise speed. Therefore, the transportation planner
determine the optimal signal timing and speed limit to minimize the should set the highest-level red time and speed limit. The optimal
overall delay and emission increment with and without traffic sig- joint decision of the signal timing and speed limit is one of the
nal control. The nonlinear and nonconvex objectives introduce the Pareto optimal solutions, which results in the tradeoff between the
complexity of the theoretical analysis. overall delay and traffic increment for the intersection.
The properties of the objective functions were numerically in- The current paper only considers one isolated intersection,
vestigated by adopting the emission rate function proposed by Nie which is not suitable for complicated urban network and must be
and Li (2013). As shown in the numerical example, the overall de- extended to cases with multiple phases and multiple movements.
lay is strictly increasing both in the red time and speed limit, Furthermore, the traffic flow in front of the intersection is assumed
namely, the shorter the red time is or the lower the speed limit is, to be uniform and stable. However, in reality, the driving behavior
the smaller the overall delay is. Thus, the transportation planner can of an individual traveler is random with respect to acceleration and
set the lowest-level red time and speed limit in terms of the overall cruise speed, which results in the nonuniform traffic flow. Finally,
delay. However, the emission increment is strictly decreasing in red in the urban network, the cooperation among the intersections is
-500
-1500
-2000
-2500
0 2000 4000 6000 8000 10000 12000
Overall delay (vehicle second)
Fig. 8. Pareto frontier in the sense of the overall delay and emission increment
important but is not investigated in the current paper. The new mod- Ge, X., Li, Z. C., Lam, W. H. K., and Choi, K. (2014). “Energy sustainable
eling methodology must be developed to study the joint design of traffic signal timings for a congested road network with heterogeneous
the speed limit and signal time in the network environment. users.” IEEE Trans. Intell. Transp. Syst., 15(3), 1016–1025.
Guensler, R. (1993). “Vehicle emission rates and average vehicle operating
speeds.” Ph.D. dissertation, Univ. of California, Davis, CA.
Guo, R., and Zhang, Y. (2014). “Exploration of correlation between envi-
Acknowledgments ronmental factors and mobility at signalized intersections.” Transp. Res.
Part D, 32, 24–34.
The work described in this paper was supported by the National Guo, X., and Yang, H. (2009). “User heterogeneity and bi-criteria system
Natural Science Foundation of China (71471068), the Major optimum.” Transp. Res. Part B, 43(4), 379–390.
Program of National Social Science Foundation of China Han, K., Liu, H., Gayah, V., Friesz, T., and Yao, T. (2016). “A robust
(13&ZD175), and Huazhong University of Science and Technol- optimization approach for dynamic traffic signal control with emission
ogy (5001300001). considerations.” Transp. Res. Part C, 70, 3–26.
Hao, P., Ban, X., and Whon Yu, J. (2015). “Kinematic equation-based
vehicle queue location estimation method for signalized intersections
References using mobile sensor data.” J. Intell. Transp. Syst., 19(3), 256–272.
Hassel, D., and Weber, F. J. (1993). “Mean emissions and fuel consumption
Andre, M., and Hammarström, U. (2000). “Driving speeds in Europe for of vehicles in use with different emission reduction concepts.” Sci. Total
pollutant emissions estimation.” Transp. Res. Part D, 5(5), 321–335. Environ., 134(1–3), 189–195.
Andre, M., and Pronello, C. (1997). “Relative influence of acceleration and Izadpanah, P., Hellinga, B., and Fu, L. (2009). “Automatic traffic shockwave
speed on emission under actual driving conditions.” Int. J. Veh. Des., identification using vehicles’ trajectories.” Proc., 88th Annual Meeting of
18(3–4), 340–353. the Transportation Research Board (CD-ROM), Washington, DC.
Bokare, P. S. (2013). “Study of effect of speed, acceleration and deceler- Jensen, S. S. (1995). “Driving patterns and emissions from different types
ation of small petrol car on its tail pipe emission.” Int. J. Traffic Transp. of roads.” Sci. Total Environ., 169(1–3), 123–128.
Eng., 3(4), 465–478. Liao, T. Y. (2013). “A fuel-based signal optimization model.” Transp. Res.
Chang, T. H., and Lin, J. T. (2000). “Optimal signal timing for an over- Part D, 23, 1–8.
saturated intersection.” Transp. Res. Part B, 34(6), 471–491. Liao, T. Y., and Machemehl, R. B. (1997). “Development of an aggre-
Cheng, Y., Qin, X., Jin, J., and Ran, B. (2010). “An exploratory shockwave gate fuel consumption model for signalized intersections.” Transp. Res.
approach for signalized intersection performance measurements using Rec., 1641, 9–18.
probe trajectories.” Proc., 89th Annual Meeting of Transportation Li, X., Li, G., Pang, S. S., Yang, X., and Tian, J. (2004). “Signal timing of
Research Board (CD-ROM), Washington, DC. intersections using integrated optimization of traffic quality, emissions
Coelho, M. C., Farias, T. L., and Rouphail, N. M. (2005). “Impact of speed and fuel consumption: A note.” Transp. Res. Part D, 9(5), 401–407.
control traffic signals on pollutant emissions.” Transp. Res. Part D, Li, Z. C., and Ge, X. (2014). “Traffic signal timing problems with environ-
10(4), 323–340. mental and equity considerations.” J. Adv. Transp., 48(8), 1066–1086.
Dai, S., Bi, X., Chan, L. Y., He, J., Wang, B., and Wang, X. (2015). Miettinen, K. (2012). Nonlinear multiobjective optimization, Vol. 12,
“Chemical and stable carbon isotopic composition of pm2.5 from Kluwer Academic Publishers, Dordrecht, Netherlands.
on-road vehicle emissions in the PRD region and implications for ve- Miller, A. J. (1963). “Settings for fixed-cycle traffic signals.” J. Oper. Res.
hicle emission control policy.” Atmos. Chem. Phys., 15(6), 3097–3108. Soc., 14(4), 373–386.
Dion, F., Rakha, H., and Kang, Y. S. (2004). “Comparison of delay esti- Ministry Environmental Protection of People’s Republic of China. (2014).
mates at under-saturated and over-saturated pre-timed signalized inter- “Environment statistics annual report 2013.” China Environmental
sections.” Transp. Res. Part B, 38(2), 99–122. Science Press, Beijing.