Optimal Cycle-Length Formulas

for Intersections With or Without

Transit Signal Priority
Bart Wolput, Eleni Christofa, and Chris M. J. Tampère

Webster’s formula for optimal green split and cycle length is a common studies and traffic signal manuals for determining green splits (4, 5,
way to optimize signal settings at signalized intersections. However, these 8, 9, 12–18).
equations are intended to be used for optimizing signal timing plans
for fixed-time isolated intersections, balancing phase utilization and mini- λi
yi = (1)
mizing car delay for undersaturated conditions. These formulas are less si
appropriate for near-saturated and oversaturated conditions because of
the asymptotical behavior of the cycle-length formula when the inter- P

section flow ratio approaches one. These formulas were not conceived for Y = ∑ yi (2)
i =1
operation under transit signal priority (TSP) strategies. The main objec-
tive of this paper is to develop formulas for the optimal cycle length that yi
Gi = i (C − L ) (3)
can be applied when mild TSP strategies, such as green extension or green Y
extension with phase rotation, are implemented or when the intersection
operates under near-saturated or oversaturated conditions. A formula where
for optimal cycle length is developed as a function of lost time and inter­
section flow ratio. The formula is applicable both with and without TSP. C = cycle length (s),
It appears that for green split, no new formulas are required because the L = total lost time (s),
Webster equisaturation approach is close to optimal. The cycle-length Gi = green time for phase i (s),
formula is validated for undersaturated, near-saturated, and oversatu-
λi = arrival flow rate for lane group i (passenger car equivalents
rated conditions with and without TSP for two-, three-, and four-phase
per second),
si = saturation flow rate for lane group i (passenger car equivalents
signalized intersections. Compared with Webster’s formula, the proposed
per second),
formula shows a significant improvement in reducing person delay. Com-
P = total number of phases in a cycle,
pared with Webster’s formula, average person delay is 6%, 17%, and
yi = critical lane group flow ratio of phase i, and
22% less and average bus delay is 16%, 27%, and 35% less for two-,
Y = sum of flow ratios for all critical lane groups: intersection
three- and four-phase signalized intersections, respectively.
flow ratio.
In addition to the formulas for determining the green splits, ana-
When optimizing signal settings such as cycle lengths and green lytical formulas have been developed for calculating the optimal cycle
splits, the practitioner can choose between optimization based on length. These formulas approximate the optimal cycle length given
analytical formulas (1–8) or simulation-based optimization (9–11). a specific objective function. Webster developed a regression formula
Webster (4) developed the critical lane analysis method, which ana- for minimizing vehicle delay (4). Akçelik adapted the coefficients of
lyzes and identifies critical ratios of demand over capacity for each that formula to develop four optimal cycle length equations that mini-
lane group. With Equations 1 through 3, the green splits can be calcu- mize delay, queue length, and fuel consumption and a performance
lated on the basis of equisaturation. Equation 1 calculates the critical index that combines the operational cost of stops and delay (5). In the
flow ratio for phase i. The ratio of the phase flow ratio over the sum of Highway Capacity Manual, a method is described for optimizing cycle
the critical flow ratios of all phases, which is calculated in Equation 2, lengths with a variable asymptotical value for user preference criti-
is used for allocating green time to a phase (Equation 3). Because cal intersection flow ratio (Yc), which is an operational measure (8).
of its simplicity, this method is commonly used in various research This method, however, indicates only at which default value the cycle
length goes to infinity. Depending on the βY (i.e., the coefficient equal
to 1 over Yc), the optimal cycle-length formula is modified according
B. Wolput and C. M. J. Tampère, Leuven Mobility Research Centre, Centrum for
Industrial Policy, University of Leuven, Celestijnenlaan 300A, P.O. Box 2422,
to user preferences of desired traffic demand. However, this param-
3001 Heverlee, Belgium. E. Christofa, Department of Civil and Environmental eter significantly influences the optimal cycle length. All these optimal
Engineering, College of Engineering, University of Massachusetts, Amherst, cycle-length formulas have the same general form, which can be found
216 Marston Hall, 130 Natural Resources Road, Amherst, MA 01003-9232. in Equation 4. However, the coefficients differ as a result the objective
Corresponding author: B. Wolput, [email protected]. functions used.
Transportation Research Record: Journal of the Transportation Research Board,
No. 2558, Transportation Research Board, Washington, D.C., 2016, pp. 78–91. β L i L + β1
Copt = (4)
DOI: 10.3141/2558-08 β 0 − βY i Y

78
Wolput, Christofa, and Tampère 79

where • Validate the obtained Copt formula in Vissim (i.e., verify if it per-
forms better than Webster’s formula for two-, three- and four-phase
Copt = optimal cycle length (s) and
intersections with PR + GE, GE, or no TSP).
βi = regression coefficients with the following values:
• Webster (4) minimal delay: β0 = 1, β1 = 5, βY = 1, and
The CAPACITEL model is a stochastic model that requires multiple
βL = 1.5;
• Akçelik (5) minimal delay: β0 = 1, β1 = 4, βY = 1, and seeds to be run. Although a 1-h simulation takes only 0.5 s, the used
βL = 1.6; heuristic (i.e., grid search) is too slow to perform real-time minimiza-
• Akçelik (5) minimal queue length: β0 = 1, β1 = 6, βY = 1, tion of delay heuristically: 30 to 45 min. The model, however, could be
and βL = 1.1; used in combination with a faster optimization heuristic when lon-
• Akçelik (5) minimal fuel consumption: β0 = 1, β1 = 6, ger optimization horizons are used. For shorter horizons, analyti-
βY = 1, and βL = 1.8; cal models based on kinematic wave theory are more appropriate
• Akçelik (5) minimal performance index: β0 = 1, β1 = 6, (2, 3, 7). The signal control module used in this research is presented
βY = 1, and βL = 1.6; and elsewhere (21).
• Highway Capacity Manual (8) operational optimum:
β0 = 1, β1 = 0, βY = 1/Yc, and βL = 1.
Objective Function
Signalized intersections are responsible for a high percentage of
transit delay. Consequently, several research efforts have focused The objective function used to optimize signal settings was weighted
on providing priority to buses at signalized intersections to enhance by a passenger occupancy function of car and transit vehicle delay.
regularity and commercial speed (1–3, 7, 11–13, 16, 19). Webster’s The weights used are those derived in previous research (20). The
formula is still considered a relevant initial estimate in many signal calculation of the objective function is as follows:
timing manuals and is extensively used in practice (8, 12, 18, 19).
The obtained green splits and cycle lengths are then usually modi-
fied to accommodate pedestrians and vehicle actuation. However, OF =
(d c i wc ,i i Nc + dT i wT,i i N T )
(5)
a first potential problem is that there is no standard methodology to ( wc ,i i Nc + wT,i i N T )
account for the existence of transit signal priority (TSP) strategies. A
second problem with Webster’s formula is in the cycle length estima- where
tion. Webster’s formula is suitable only for undersaturated conditions
OF
_ = objective function value of person delay (s);
because of the significant and unbounded increase of the optimal cycle
dc = average car delay (s);
length at intersection flow ratios close to one.
wc,i = passenger occupancy for cars: 1.2 persons per vehicle;
In previous work, formulas were proposed for TSP strategies that
N
_c = number of cars;
caused a significant capacity shift because of red truncation (1).
dT = average bus delay (s);
This shift was demonstrated in research on the effectiveness of TSP
wT,i = passenger occupancy for buses: 30 persons per vehicle; and
strategies (20). A need remained to develop simplified analytical
NT = number of buses.
formulas that optimize cycle lengths and splits for a wide range of
traffic demands and which are valid for TSP strategies with limited
capacity shift, such as green extension (GE) or phase rotation with
green extension (PR + GE) (20). Test Site

Figure 1 shows the test intersection and three potential phase designs.
It consists of four approaches with three lanes per approach, one
Research Objective
each for left-turning, through, and right-turning movements. No bus
The contribution of this paper is to develop and validate a formula stops were considered in this study. The four-phase intersection was
for estimating the optimal cycle length (in minimizing person used for the sensitivity analysis of the design variables for the opti-
delay) for undersaturated, near-saturated, and oversaturated con- mal cycle-length formula. It is assumed that if there are no effects
ditions at isolated intersections when TSP strategies with limited of transit variables for four-phase intersections, there will also be no
capacity shift are in place. These strategies include green exten- effects for two- or three-phase intersections. This assumption is vali-
sion or phase rotation with green extension but exclude more intru- dated with Vissim in the last part of this paper. The three- and four-
sive strategies that use red truncation. The following methodology phase intersections were used to perform the optimization of various
was used: demand scenarios in CAPACITEL, and the regression analysis for the
optimal cycle length was performed. The two-, three-, and four-phase
• Conduct sensitivity analyses of various traffic and transit vari- intersections were used for the validation in Vissim.
ables with traffic and signal timing optimization model CAPACITEL
to determine which variables should be included and for which
boundary conditions the conclusions are valid (1). Sensitivity Analysis of Transit
• Optimize signal settings (G, C) by minimizing total person delay and Traffic Parameters
using CAPACITEL for 20,000 traffic scenarios for three- and four-
phase intersections when green extension or a combination of phase The impact of six variables on optimal green splits and cycle lengths
rotation and green extension is implemented. was evaluated through sensitivity analysis. The variables were inter-
• Determine the coefficients of a regression equation for Copt with section flow ratio (Y), flow ratio of the first transit phase as a percent-
linear regression analysis. age of Y (yT = y1), transit frequency in Phase 1 (FT,1), transit frequency
80 Transportation Research Record 2558

Phase 1 Phase 2

Phase 1 Phase 2 Phase 3

Phase 1 Phase 2 Phase 3 Phase 4

d2 d3 d4

d1 d1
d3 d4
d2

Transit Line 1 Transit Line 2 Transit Line 3 Transit Line 4

FT,2 FT,4 FT,4
FT,2 FT,3
FT,1 2 2
2 2

FIGURE 1   Test intersection layout and phase design with car [d i (vph)] and bus demand [F T,i (buses per hour)].

in Phase 3 (FT,3), distance of the priority detector from the intersection Sensitivity Analysis 5: FT,3 (buses per hour): 0, 3, 6, 9, 12, 15, 18,
stop bar (Ddet), and total lost time (L). With an optimization algorithm 21, 24, 27; and
based on grid search, the solution space of the signal settings (C, Gi) Sensitivity Analysis 6: Ddet (s): 10, 15, 20, 25, 30, 35, 40, 50, 60, 70.
was scanned for obtaining the optimal signal settings for a certain
combination of the variables. Sensitivity Analyses 2 through 6 were The lost time (L) and saturation flow (s) are used as a direct input
performed for three Ys: unsaturated (0.6), near-saturated (0.75), and in CAPACITEL. The lost time (L) is equal to the cycle length minus
oversaturated (0.875). The four-phase intersection base case settings the sum of the effective green lengths (19). It is assumed that yellow
were as follows: indicates to the driver to stop, if possible; so it is not considered as
effective green time. The saturation flow (s) is the outflow constraint,
• Intersection flow ratio (Y): 0.6, 0.75, and 0.875; which is assumed to be approximately 1,800 vph per lane in an urban
• Total lost time (L): 20 s; environment, as long as no conflicting flows are present.
• Critical flow ratios (y1, y2, y3, y4): 35, 15, 35, 15 (as a percentage
of Y);
• Bus frequency for Phases 1 (FT,1) and 3 (FT,3) (i.e., conflicting Optimal Cycle Length
phases): 15 buses per hour for both; Figure 2 shows the impact of the six variables. The intersec-
• Detector location (Ddet): 20 s, convertible to meters with free-
tion flow ratio (Y ) and total lost time (L) demonstrate the impact
flow speed of 10 m/s; on the optimal cycle length as also described by Webster (4).
• Transit green extension beyond normal green (GT): Ddet + 10 s; At higher intersection flow ratios (Figure 2a) or total lost times
• Minimum green length (Gi,min): 10 s (22); (Figure 2c), the optimal cycle length increases. The difference
• Saturation flow per lane (s): 1,800 vehicles per hour (vph); in the optimal cycle length between L = 16 and L = 28 is greater
and when Y increases (Figure 2a). The same trend can be distilled for
• TSP strategy: phase rotation, green extension. the differences observed between Y = 0.6 and Y = 0.875 when L
gets higher (Figure 2c). This effect illustrates that the correlation
For the six sensitivity analyses, the following variables were between the two variables has to be taken into account; this cor-
changed: relation is omitted in Webster’s equation (Equation 4). Figure 2a
and Figure 2c clearly contradict the inverse trend (x/1 − Y ) as
Sensitivity Analysis 1: predicted by Webster and indicate a more linear evolution. For Y = 0.9
– Y: 0.3, 0.6, 0.7, 0.75, 0.775, 0.8, 0.825, 0.85, 0.875, 0.9 and and L = 28 (Figure 2a), the degree of saturation (Y • C/C − L) is 103%,
– L (s): 16, 20, 28; and the optimal cycle length is 218 s; Webster’s formula would
Sensitivity Analysis 2: y1 (%): 10, 15, 20, 25, 30, 35, 40, 45, 55, 60; have estimated it to be 470 s.
Sensitivity Analysis 3: L (s): 16, 20, 24, 28, 32, 36, 40, 44, 48, 52; Other variables show no clear trend for the impact they have on
Sensitivity Analysis 4: the optimal cycle length. Only for detector distances above 30 s of
– FT,1 (buses per hour): 0, 5, 10, 15, 20, 25, 30, 40, 50, 60 and free-flow travel time (Figure 2f ), there is a constant rise in optimal
– FT,3 (buses per hour): 0; cycle length for near-saturated conditions (Y = 0.75). The farther
 Wolput, Christofa, and Tampère 81

Cycle Time (s)

Cycle Time (s)
Cycle Time (s)

yT (%) L (s)
Y

(a) (b) (c)

Cycle Time (s)

Cycle Time (s)

Cycle Time (s)

FT,1 (vph) FT,3 (vph) Ddet (s)

(d) (e) (f)

FIGURE 2   Optimal cycle lengths for sensitivity analyses.

the detector, the higher the impact of the buses on the capacity split split formula can be used in addition to the proposed cycle-length
because of the green extension; the result is nonpriority flows going formula. One particular phenomenon occurs for very low intersection
into temporary oversaturation. Previous work from Liu et al. demon- flow ratios. With Y equal to 0.3, the shorter phases (yi = 15%) have
strated that the optimal detector distance is between 15 and 30 s (23). a bigger part of the green split as compared with higher intersection
So it is a valid assumption to limit the variables for the regression flow ratios—although the phase flow ratios are the same—because
analysis to Y and L as long as the detector distance is limited to 30 s of the chosen minimum green of 10 s for each phase and the low opti-
of free-flow travel time from the stop bar. mal cycle length (Figure 2). When the green lengths are below the
minimum green length, they can be set equal to the minimum green.

Optimal Green Split

Sensitivity Analysis Conclusion
Figure 3 shows the impact of the parameters on the green splits. The
The sensitivity analysis suggests that for mild TSP variants, it does
optimal green values are presented as their percentage of the total green.
not matter whether the TSP is there. The sensitivity of cycle length
The base case values for y1, y2, y3, and y4 are according to Webster’s
with respect to bus frequency appears to be negligible. Green dis-
proportional green split (i.e., equisaturation) approach (Equation 3)
tributions are comparable with the ones expected from Webster’s
equal to 35, 15, 35, and 15. They are used for the sensitivity analysis
equisaturation approach for detector ranges below 30 s of free-flow
of Y (Figure 3a, left), L (Figure 3a, right), FT,1 (Figure 3b, left), FT,3
travel time from the stop bar.
(Figure 3b, center), and Ddet (Figure 3b, right).
Optimal cycle lengths around 110 to 140 s for oversaturated con-
In general, green splits have a constant trend over a variety of param-
ditions show a strong similarity with the work of Denney et al. (24)
eter values (Figure 3) and find a close alignment to the phase flow ratios
and Day et al. (25). These studies maximized throughput, whereas
(yi). This trend is in agreement with the optimal green splits presented
in this paper minimizing person delay is used as the objective func-
by Webster’s proportional green split formula, which is equal to yi
tion. The results confirm that the objective of maximizing throughput
when expressed in a percentage. For the equisaturation approach
aligns with minimizing person delay for oversaturated intersections,
of Webster’s green split formula to be valid, the ratio of the phase criti-
which is an equivalence used in other traffic control literature, such
cal flow ratio over the intersection flow ratio should be equal to the
as work by Stoilov et al. (26). The implication is that the asymptoti-
ratio of the phase green over the total green. This result implies that
cal behavior for Y → 1 is the cause of Webster’s deviant cycle length
a new regression for estimating green lengths is unnecessary and the
estimations at high intersection flow ratios, not the objective function
equisaturation approach can be used. The validation results presented
used for its regression.
later in this paper confirm this conclusion.
With regard to boundary deviations, the following conclusions are
drawn for the validity of the new formulas in the solution space. Framework for Regression Analysis
When the detector distance is far from the stop bar, the green splits
are favored in the direction of phases without transit vehicles to com- The signal settings were optimized for more than 20,000 input param-
pensate for the additional green allocated to the transit phases. When eter combinations (10,000 for three-phase intersections, 10,000
the detector distance in the regression is limited, Webster’s green for four-phase intersections). With these results, a new regression
82 Transportation Research Record 2558

G1 (%)

G1 (%)

G1 (%)
yT (%) L (s)
Y
G2 (%) L = 16 L = 20 L = 28 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875

G2 (%)

G2 (%)
yT (%) L (s)
Y
L = 16 L = 20 L = 28 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875
G3 (%)

G3 (%)

G3 (%)
yT (%) L (s)
Y
L = 16 L = 20 L = 28 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875
G4 (%)

G4 (%)

G4 (%)
yT (%) L (s)
Y
L = 16 L = 20 L = 28 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875

(a)
G1 (%)

G1 (%)

G1 (%)

FT,1 (vph) FT,3 (vph) Ddet (s)

Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875
G2 (%)

G2 (%)

G2 (%)

FT,1 (vph) FT,3 (vph) Ddet (s)

Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875
G3 (%)

G3 (%)

G3 (%)

FT,1 (vph) FT,3 (vph) Ddet (s)

Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875
G4 (%)

G4 (%)

G4 (%)

FT,1 (vph) FT,3 (vph) Ddet (s)

Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875 Y = 0.6 Y = 0.75 Y = 0.875

(b)

FIGURE 3   Optimal green splits for six sensitivity analyses.

Wolput, Christofa, and Tampère 83

equation was calibrated for approximating optimal cycle length. One 1

Copt = (6)
hour of simulation and five seeds were used for each combination. β 0 + βY i Y + β L i L + βY & L i Y i L
The input parameters that were combined are intersection flow ratio
(Y), flow ratio of the phases (y1, y2, y3, y4) expressed as a percentage where βi = coefficient i, to be determined by linear regression
of Y, lost time as the sum of yellow and all red times (L), detector analysis.
distance expressed in time from the stop bar at a user-defined free- The regression analysis resulted in Equation 7 and a high adjusted
flow speed (Ddet), and the frequency of buses for all phases (FT,1, FT,2, R2 value (.87) of explained variance. All coefficients are statistically
FT,3, FT,4). The values of the variables for the 20,000 combinations significant with a P-value < .0005 and a low standard error of 0.001.
were sampled randomly for three- and four-phase intersections from This equation can be directly used to estimate optimal cycle lengths
a certain range. Those ranges are presented here. after calculating the intersection flow ratio (Y) and the total lost time
(L). The signs of the coefficients imply that higher Y and L values
• Three phase:
will result in a lower denominator and thus a higher cycle length.
– Y: Y ∈ ℝ: 0.6 ≤ Y ≤ 0.925 Having a TSP or no TSP does not affect the calculated cycle lengths
– yi (%): yi ∈ ℝ: 10 ≤ yi ≤ 50; ∑ i=1 yi = 100
P
because only Y and L are used as inputs.
– L (s): L ∈ ℕ: 12 ≤ yi ≤ 24
– FT,i (buses per hour): FT,i ∈ ℝ: 0 ≤ FT,i ≤ 30; ∑ i=1 FT,i ≤ 45
P
100
– Ddet (s): Ddet ∈ ℝ: 10 ≤ Ddet ≤ 30 Copt = (7)
• Four phase: 5.146 − 4.625 i Y − 0.1045 i L + 0.09483 i Y i L
– Y: Y ∈ ℝ: 0.6 ≤ Y ≤ 0.9
– yi (%): yi ∈ ℝ: 5 ≤ yi ≤ 45; ∑ i=1 yi = 100
P

Inverse Linear Regression

– L (s): L ∈ ℕ: 16 ≤ yi ≤ 32
– FT,i (buses per hour): FT,i ∈ ℝ: 0 ≤ FT,i ≤ 30; ∑ i=1 FT,i ≤ 45
P with Transformation of C/L
– Ddet (s): Ddet ∈ ℝ: 10 ≤ Ddet ≤ 30
Figure 4 shows the results from the sensitivity analysis (Copt) for Fig-
ure 2a (L = 20) and Figure 2c (Y = 0.875) compared with the estimation
Other fixed parameters are as follows:
of the optimal cycle length by Webster’s equation (Equation 4), Equa-
• tion 7, and Equation 8, which is discussed later in this section. When
Transit green extension beyond Gi(GT): Ddet + 10 s;
• approaching saturation (i.e., with high values for L or Y or both),
Minimum green for a phase i (Gi,min): 10 s;
• Saturation flow per lane (s): 1,800 vph; and Webster’s formula estimates very high values for the optimal cycle
• TSP strategy: phase rotation with green extension. length. Even though Equation 7 performs well within its preset validity
range (L ≤ 30), at values of L exceeding this boundary, it diverges
quickly to overestimation of the optimal cycle length value. The over-
Regression Analysis Results estimation is even greater than that from Webster’s formula (Figure 4b),
compared with the optimized cycle length value obtained from the
Inverse Linear Regression simulation-based optimization. Although such a high value for L
normally does not occur and is in principle outside the validity
After the cycle lengths of the 20,000 scenarios were optimized, a range, the authors attempted to find an alternative form of Equation 7
regression was performed to calibrate the coefficients of the new that would not exhibit this undesirable behavior. In this alterna-
formula to approximate the optimal cycle length. First, a transfor- tive equation (Equation 8), L is in the numerator instead of the
mation was done by taking the inverse formulation for the linear denominator; the formula therefore does not reach very high values
regression (Equation 3). at high values of L. The regression analysis (for L ≤ 30 and the

400 1,600
350 1,400
300 1,200
250 1,000
Copt (s)

Copt (s)

200
800
150
600
100
400
50
0 200
0
3

7
75

5
85

9
0.

0.

0.

77

0.
82

87

0.

16 20 24 28 32 36 40 44 48
0.

0.
0.

0.

0.

Y L (s)
(a) (b)
Eq. 7 Eq. 8 Webster Simulation Eq. 7 Eq. 8 Webster Simulation

FIGURE 4   Comparison of cycle lengths: sensitivity regarding (a) Y (L 5 20 s) and (b) L (Y 5 0.875)
(eq. 5 equation).
84 Transportation Research Record 2558

full range of Y ) of Equation 8 obtained almost equal optimal cycle comparable length, as would be expected from Figure 4. However, of
length estimations as those of Equation 7, as well as an almost iden- the 160 comparisons over 60 demand scenarios, Webster’s formula
tical adjusted R2 value (.86) of explained variance. All coefficients had an absolute lower person delay in only two cases—Cases 2
are statistically significant with a P-value < .0005 and a standard and 46—both with no priority. The extra delay, however, was for
error equal to 0.02. intersection flow ratios that indicate undersaturated traffic conditions;
the delay was less than 1 s per person and not statistically signifi-
1.7 i L cant. In all other cases, the performance of the new formula was
Copt = (8)
YiL better. For scenarios of the two-, three-, and four-phase inter-
1−Y +
180 sections, the average person delay of the new formula signal set-
tings was lower than the one obtained with Webster’s settings
For further validating the formula for more TSP strategies and (Figure 6, a through c). For the two phases, the new formula per-
demand scenarios in the next sections of this paper, Equation 7 is formed significantly better in 50% of tested cases, in 90% of the
used. (Equation 8 had not yet been developed at the time the exten- tested cases for three phases, and in 95% of the tested cases for
sive validation study was performed). However, because the cal- four phases. In the other cases, the new formula performed better in
culated optimal cycle lengths of both formulas are almost identical person delay, but the differences were not statistically significant.
within their validity range, it is reasonable to assume that Equation 8 Especially when high intersection flow ratios were present, the new
will deliver similar validation results. formula for optimal cycle length always performed significantly
better for person delay.
Smaller queues as a result of shorter and improved cycle length
Framework for Validation estimates have an impact on bus delay as well, showing reductions
in bus delay. Only in six of the 160 cases (Cases 2, 21, 25, 42, 43,
For proper validation, the person delays for 60 demand scenario and 46) did the bus delay increase with the new formula (Figure 6,
cases of the new formula (Equation 7) were compared with those d through e). The six cases were all with no priority and low inter-
obtained from Webster’s optimal cycle-length formula. The inter- section flow ratios. Thus the impact was limited between 0.3 and
section used for validation is shown in Figure 1. Figure 5 shows the 5.4 s of extra average bus delay when the new formula was used.
transit and traffic demands for two, three, and four phases for the The average bus delay differences for those six cases were all
tested cases with randomly chosen exogenous variables within not statistically significant, with high P-values. In 81% of the
the boundaries determined by the sensitivity analysis. The validation 160 cases, the new formula had statistically significant lower average
is set up for illustrating the applicability of the formula for various bus delays.
traffic and transit demands, number of phases, and TSP strategies, In general, the highest gains (both for buses and other traffic) with
as follows: the new formula for optimal cycle lengths were obtained at higher
intersection flow ratios and intersections with multiple phases. The
• Four phases, 20 demand cases (Cases 1 through 20): more phases, and the longer the lost time (L), the higher the impact
– No priority, on reducing person delay. Whereas the gains in bus and person delay
– Green extension, and are limited for two-phase intersections, they rapidly increase for
– Phase rotation and green extension; three- and four-phase intersections at higher intersection flow ratios.
• Three phases, 20 demand cases (Cases 21 through 40): The general results are summarized in Table 1 and show the poten-
– No priority, tial of the new formula in reducing person and bus delay. The new
– Green extension, and formula never performs significantly worse than Webster’s formula,
– Phase rotation and green extension; and but mostly performs significantly better.
• Two phases, 20 demand cases (Cases 41 through 60):
– No priority and
– Green extension. Conclusion

The microsimulation software Vissim was used, and every scenario The main objective of this paper is to develop an optimal cycle-length
was run 10 times for proper comparison for a 1-h simulation plus a formula that is applicable to unsaturated, near-saturated, and over-
10-min warmup. To investigate the significance of the improvement, saturated conditions at isolated intersections. Research finds that for
a t-test was conducted with a significance level of 95%. An assumption mild TSP variants, such as green extension or phase rotation with
was made that the results follow a normal distribution. green extension, it does not matter whether TSP is implemented. The
Total lost time (L) was introduced in Vissim by changing the inter- old cycle formulas of Webster and Akçelik are not valid for intersec-
phase duration. The saturation flow of each lane was calibrated by tions with high flow ratios because of their asymptotical formulation.
applying reduced speed areas. The average speed of the speed areas The optimal cycle lengths for minimizing person delay align perfectly
was reduced until an average output of 1,800 vph was found over five with other research that used throughput maximization as the objective
random seeds. function for oversaturated intersections (24–26).
In addition—and as confirmed by both the sensitivity analysis and
the validation results in this paper—neither the frequency of buses
Results of Validation nor the presence of conflicting lines have any impact on optimal
signal settings for the considered range of 0 to 60 vph and detector
The validation results are shown in Figure 6. At low intersection flow distances below 35 s of free-flow travel time from the stop bar. The
ratios, the gains of the new formula were limited because Webster’s difference with the authors’ previous work (1) is that the formula is
formula and the new formula estimate the optimal cycle length at a validated only for TSP strategies with negligible capacity shift: green
 278
366

184
493

127
179

451
265
10
12

10

10

10
12
1

4
0 0 1 0
173 10 300 10 424 10 119 10
6 Case 1 7 14 Case 2 13 10 Case 3 9 2 Case 4 1
225 Y = 0.6 225 146 Y = 0.64 146 396 Y = 0.64 396 318 Y = 0.66 318
10 173 10 300 10 424 10 119
0 0 1 0
366

278

493

184

179

127

265

451
12

10

10

10

12

10
0

4
554
353

154

499
292

267
445
311
10

10
13

10

10
11
0

4
0 6 0 0
126 10 585 10 250 10 373 10
3 Case 5 3 0 Case 6 0 0 Case 7 0 4 Case 8 3
221 Y = 0.7 221 207 Y = 0.72 207 250 Y = 0.73 250 219 Y = 0.74 219
10 126 10 585 10 250 10 373
0 5 0 0
353

554

154

292

499

445

267
311
10

14

10

12

10

10
0

6
179
133

144
314

199
372

576
611
10

10

10

10
11

0
0 0 4 6
693 10 621 10 269 10 92 10
4 Case 9 5 10 Case 10 10 10 Case 11 10 10 Case 12 9
297 Y = 0.74 297 237 Y = 0.75 237 578 Y = 0.81 578 158 Y = 0.82 158
10 693 10 621 10 269 10 92
0 0 4 5
133

179

314

144

372

199

576

611
10

10

10

10

10
0

4
209
215

554
553

566
258

468
329
10

12

10

10

10
0

0
4 1 0 8
531 10 88 10 107 10 125 10
2 Case 13 2 10 Case 14 9 8 Case 15 9 0 Case 16 0
504 Y = 0.83 504 265 Y = 0.83 265 564 Y = 0.85 564 611 Y = 0.87 611
10 531 10 88 10 107 10 125
3 1 0 8
215

209

553

554

258

566

329

468
12

10

10

10

10
0

8
341
523

238
329

593
224

587
258
10

10

10

10
11

0 3 7 0
143 10 786 10 229 10 113 10
8 Case 17 8 14 Case 18 14 1 Case 19 1 12 Case 20 12
538 Y = 0.88 538 194 Y = 0.88 194 549 Y = 0.9 549 624 Y = 0.9 624
10 143 10 786 10 229 10 113
0 3 7 0
523

341

329

238

224

593

258

587
12

10

10

10

10
0

(a)

FIGURE 5   Transit (red dotted lines) and traffic (full black lines) randomly generated demands used for validation:
(a) four-phase intersection.
(continued on next page)
 289

230

426

416
10
12

59

10

55

10

52

10
12

48
1

4
0 0 1 0
389 10 576 10 482 10 630 10
6 Case 21 7 14 Case 22 13 10 Case 23 9 2 Case 24 1
365 Y = 0.6 365 319 Y = 0.64 319 223 Y = 0.64 223 116 Y = 0.66 116
10 389 10 576 10 482 10 630
0 0 1 0
289

230

426

416
59
12

10

55

10

52

10

48
12

10
0

4
441

284

499

573
11
10

52

10
13

50

10

26

10

40
0

4
0 6 0 0
328 10 818 10 289 10 440 10
3 Case 25 3 0 Case 26 0 0 Case 27 0 4 Case 28 3
485 Y = 0.7 485 156 Y = 0.72 156 526 Y = 0.73 526 299 Y = 0.74 299
10 328 10 818 10 289 10 440
0 5 0 0
441

284

499

573
52

10

50
14

10

12
26

10

40

10
0

6
139

306

534

449
11
10

48

10

37

10

31

10

37
0

0
0 0 4 6
626 10 459 10 502 10 830 10
4 Case 29 5 10 Case 30 10 10 Case 31 10 10 Case 32 9
537 Y = 0.74 537 561 Y = 0.75 561 388 Y = 0.81 388 158 Y = 0.82 158
10 626 10 459 10 502 10 830
0 0 4 5
139

306

534

449
48
10

10

37

10

31

10

37

10
0

4
299

479

306

594
11
10

12

10

35

10

15

10

28
0

0
4 1 0 8
800 10 521 10 933 10 329 10
2 Case 33 2 10 Case 34 9 8 Case 35 9 0 Case 36 0
384 Y = 0.83 384 459 Y = 0.83 459 274 Y = 0.85 274 611 Y = 0.87 611
10 800 10 521 10 933 10 329
3 1 0 8
299

479

306

594
11
12

10

35

10

15

10

28

10
0

8
690

285

707

458
11
10

26

10

22

10

19

10

21
0

0 3 7 0
380 10 628 10 342 10 842 10
8 Case 37 8 14 Case 38 14 1 Case 39 1 12 Case 40 12
475 Y = 0.88 475 637 Y = 0.88 637 549 Y = 0.9 549 284 Y = 0.9 284
10 380 10 628 10 342 10 842
0 3 7 0
690

285

707

458
26
12

10

22

10

19

10

21

10
0

(b)

FIGURE 5 (continued)   Transit (red dotted lines) and traffic (full black lines) randomly generated demands used for validation:
(b) three-phase intersection.
(continued)
 386

553

518

724
10
12

74

10

70

10

67

10
12

62
1

4
0 0 1 0
74 10 70 10 67 10 62 10
6 Case 41 7 14 Case 42 13 10 Case 43 9 2 Case 44 1
657 Y = 0.6 657 572 Y = 0.64 572 615 Y = 0.64 615 437 Y = 0.66 437
10 74 10 70 10 67 10 62
0 0 1 0
386

553

1,038 518

724
74
12

10

70

10

67

10

62
12

10
0

4
781

414

440
10

39
11
0
10

66

10
13

63

10

52
0

4
0 6 0 0
66 10 63 10 39 10 52 10
3 Case 45 3 0 Case 46 0 0 Case 47 0 4 Case 48 3
473 Y = 0.7 473 855 Y = 0.72 855 276 Y = 0.73 276 872 Y = 0.74 872
10 66 10 63 10 39 10 52
0 5 0 0

1,118 1,038
781

414

440
12
39

10
66

10

63
14

10

52

10
0
0

6
392

495

803
10

60

10

49

10

41

10

46
11

0
0 0 4 6
60 10 49 10 41 10 46 10
4 Case 49 5 10 Case 50 10 10 Case 51 10 10 Case 52 9
910 Y = 0.74 910 831 Y = 0.75 831 315 Y = 0.81 315 645 Y = 0.82 645
10 60 10 49 10 41 10 46
0 0 4 5
1,118
392

495

803
60
10

10

49

41
10

10

46

10
0

4
2

4
418

479

887

328
10

20
12

10

44

10

23

10

35
0

0
4 1 0 8
20 10 44 10 23 10 35 10
2 Case 53 2 10 Case 54 9 8 Case 55 9 0 Case 56 0
1,072 Y = 0.83 1,072 982 Y = 0.83 982 626 Y = 0.85 626 1,221 Y = 0.87 1,221
10 20 10 44 10 23 10 35
3 1 0 8
418

1,220 479

887

328
12
20

10

44

10

23

10

35

10
0

8
421

367

928
29
10

33

10

10

25

10

27
11

2
0

0 3 7 0
33 10 29 10 25 10 27 10
8 Case 57 8 14 Case 58 14 1 Case 59 1 12 Case 60 12
1,124 Y = 0.88 1,124 336 Y = 0.88 336 1,245 Y = 0.9 1,245 656 Y = 0.9 656
10 33 10 29 10 25 10 27
0 3 7 0
1,220
421

367

928
33
12

10

29

10

25

10

27

10
0

(c)

FIGURE 5 (continued)   Transit (red dotted lines) and traffic (full black lines) randomly generated demands used for validation:
(c) two-phase intersection.
88 Transportation Research Record 2558

Average Person Delay (s)

Case
Webster, no TSP Wolput, no TSP Webster GE Wolput GE Webster PR + GE Wolput PR + GE
(a)
Average Person Delay (s)

Case
Webster, no TSP Wolput, no TSP Webster GE Wolput GE Webster PR + GE Wolput PR + GE
(b)
Average Person Delay (s)

Case
Webster, no TSP Wolput, no TSP Webster GE Wolput GE
(c)

FIGURE 6   Average delay comparison with and without TSP: average person delay for (a) four-phase
intersection, (b) three-phase intersection, and (c) two-phase intersection.
(continued)
Wolput, Christofa, and Tampère 89

Average Bus Delay (s)

Case
Webster, no TSP Wolput, no TSP Webster GE Wolput GE Webster PR + GE Wolput PR + GE
(d)
Average Bus Delay (s)

Case

Webster, no TSP Wolput, no TSP Webster GE Wolput GE Webster PR + GE Wolput PR + GE

(e)
Average Bus Delay (s)

Case

Webster, no TSP Wolput, no TSP Webster GE Wolput GE

(f)

FIGURE 6 (continued)   Average delay comparison with and without TSP: average bus delay for (d) four-phase
intersection, (e) three-phase intersection, and (f ) two-phase intersection.
90 Transportation Research Record 2558

TABLE 1   General Performance of New Cycle-Length Formula Compared with Webster’s Formula

Two Phases Three Phases Four Phases

Measure of Effectiveness (s) (%) (s) (%) (s) (%)

Extra person delay

  Average −4 −6 −18.8 −17 −31.2 −22
  Minimum −14.0 −20 −95.5 −51 −155.5 −59
  Maximum +0.7 +2 +0.2 +1 +0.7 +2
Extra bus delay
  Average −10.2 −16 −23.0 −27 −34.0 −35
  Minimum −42.5 −36 −115.0 −65 −169.0 −74
  Maximum +1.4 +5 +3.8 +18 +5.4 +12
Significantly worse person delay cases (total cases) 0 (40) 0 (60) 0 (60)
Significantly better person delay cases (total cases) 20 (40) 54 (60) 57 (60)
Significantly worse bus delay cases (total cases) 0 (40) 0 (60) 0 (60)
Significantly better bus delay cases (total cases) 23 (40) 51 (60) 56 (60)

extension, phase rotation with green extension, or when no active References

TSP strategy is used.
The intersection flow ratio (Y), the lost time (L), and their correla- 1. Wolput B., E. Christofa, A. Carbonez, A. Skabardonis, and C. M. J.
tion were found to be the relevant parameters in the regression analy- Tampère. Optimal Traffic Signal Settings with Transit Signal Priority.
Presented at 94th Annual Meeting of the Transportation Research Board,
sis. The signal settings of 20,000 scenarios with various traffic and Washington, D.C., 2015.
transit demands were optimized for minimal person delay. With the 2. Christofa, E., K. Aboudolas, and A. Skabardonis. Arterial Traffic Signal
results from the optimal signal settings, a new formula for optimal Optimization: A Person-Based Approach. Transportation Research Part C:
cycle length was developed through regression analysis. Emerging Technologies, Vol. 66, 2016, pp. 27–47.
Finally, the formula was validated in Vissim for two, three, and 3. Christofa, E., I. Papamichail, and A. Skabardonis. Person-Based Traffic
Responsive Signal Control Optimization. IEEE Transactions on Intelligent
four phases and compared with the results from Webster’s formula Transportation Systems, Vol. 14, No. 3, 2013, pp. 1278–1289.
for 60 randomly generated demand scenarios. The applicability was 4. Webster, F. V. Traffic Signal Settings. Road Research Technical Paper 39.
tested for scenarios with and without TSP, crossing transit lines, and Road Research Laboratory, London, 1958.
different numbers of phases with different lost times. In general, the 5. Akçelik, R. Traffic Signals: Capacity and Timing Analysis. Australian
Road Research Board, Melbourne, Victoria, 1981.
proposed formula never performs statistically significantly worse
6. SITRAFFIC Traffic Engineering Workstation. Siemens, Berlin, 2005.
than Webster’s formula. The potential gains are high, especially in 7. Han, X., P. Li, R. Sikder, Z. Qiu, and A. Kim. Development and
near-saturated and oversaturated conditions. The more phases that Evaluation of Adaptive Transit Signal Priority Control with Updated
are used, the higher the reduction in delay will be. Average reduc- Transit Delay Model. In Transportation Research Record: Journal
tion of delay per person is 4 s for two phases, 18.8 s for three phases, of the Transportation Research Board, No. 2438, Transportation
Research Board of the National Academies, Washington, D.C., 2014,
and 31.2 s for four phases.
pp. 45–54.
The results allow a recommendation of the new formula for 8. Highway Capacity Manual 2010. Transportation Research Board of the
approximating optimal cycle lengths for intersections without TSP National Academies, Washington, D.C., 2010.
or with mild TSP strategies for various traffic and transit demands. 9. SIDRA INTERSECTION User Guide. Akçelik & Associates, Greythorn,
Green splits can be obtained using the equisaturation approach. For Victoria, Australia, 2012.
10. Binning, J. C. TRANSYT 15 User Guide. Transport Research Laboratory,
more intrusive TSP strategies, the authors recommend more com-
Wokingham, Berkshire, United Kingdom, 2014.
plex dedicated regression formulas for cycle length and green splits 11. Stevanovic, J., A. Stevanovic, P. T. Martin, and T. Bauer. Stochastic
(1, 20). Other research has also demonstrated that the new formula Optimization of Traffic Control and Transit Priority Settings in VISSIM.
can be a good starting point for heuristic optimization of the network Transportation Research Part C: Emerging Technologies, Vol. 16, No. 3,
cycle length (27). 2008, pp. 332–349.
12. Koonce, P., L. Rodegerdts, and K. Lee. Traffic Signal Timing Manual.
Finally, nearside transit stops were ignored in this study but Report FHWA-HOP-08-024. FHWA, U.S. Department of Transportation,
would interfere when TSP strategies are implemented. The impact 2002.
would depend on placement of the detector with respect to the bus 13. Dion, F., and B. Hellinga. A Rule-Based Real-Time Traffic Responsive
stop, the spread of the dwell time, and the average dwell time. Signal Control System with Transit Priority: Application to an Isolated
The amount of total travel time from detector to stop bar should Intersection. Transportation Research Part B: Methodological, Vol. 36,
No. 4, 2002, pp. 325–343.
be below 30 s of free-flow travel time from the stop bar for the 14. Taale, H., and H. J. Van Zuylen. Anticipatory Control: Effects of Uncer-
formula of this paper to be valid. tainty in Travel Times. Presented at 13th Mini-EURO Conference:
Handling Uncertainty in the Analysis of Traffic and Transportation
Systems, Bari, Italy, 2002.
Acknowledgments 15. Dion, F., and H. Rakha. Integration of Transit Signal Priority Within
Adaptive Traffic Signal Control Systems. Presented at 84th Annual Meet-
ing of the Transportation Research Board, Washington, D.C., 2005.
The authors thank IWT (Institute of Scientific and Technological 16. Zhou, G., and A. Gan. Design of Transit Signal Priority at Signalized
Research) and TECHNUM, as well as Vlaamse Overheid (IWT-SBO Intersections with Queue Jumper Lanes. Journal of Public Transporta-
Urban Logistics and Mobility project) for providing funding. tion, Vol. 12, No. 4, 2009, pp. 117–132.
Wolput, Christofa, and Tampère 91

17. Kouvelas, A., K. Aboudolas, M. Papageorgiou, and E. B. Kosmatopoulos. nal of the Transportation Research Board, No. 1867, Transportation
A Hybrid Strategy for Real-Time Traffic Signal Control of Urban Research Board of the National Academies, Washington, D.C., 2004,
Road Networks. Intelligent Transportation Systems, Vol. 12, No. 3, 2011, pp. 144–150.
pp. 884–894. 24. Denney, Jr., R. W., E. Curtis, and L. Head. Long Green Times and Cycles
18. Teply, S., D. I. Allingham, D. B. Richardson, and B. W. Stephenson. at Congested Traffic Signals. In Transportation Research Record: Jour-
Canadian Capacity Guide for Signalized Intersections (J. W. Gough, nal of the Transportation Research Board, No. 2128, Transportation
ed). Institute of Transportation Engineers, District 7, Kitchener, Ontario, Research Board of the National Academies, Washington, D.C., 2009,
Canada, 2008. pp. 1–10.
19. Wilson, A. Handboek Verkeerslichtenregelingen. CROW, Ede, 25. Day, C., J. R. Sturdevant, H. Li, A. Stevens, A. M. Hainen, S. M. Remias,
Netherlands, 2006. and D. M. Bullock. Revisiting the Cycle Length: Lost Time Question
20. Wolput, B., E. Christofa, and C. M. J. Tampère. On the Effectiveness of with Critical Lane Analysis. In Transportation Research Record: Jour-
Transit Signal Priority Strategies. Working paper. 2015. https://www nal of the Transportation Research Board, No. 2355, Transportation
.researchgate.net/publication/283644095_On_the_effectiveness_of Research Board of the National Academies, Washington, D.C., 2013,
_transit_signal_priority_strategies. pp. 1–9.
21. Wolput, B. Signal Control Module in CAPACITEL for Isolated Intersec- 26. Stoilov, T., K. Stoilova, M. Papageorgiou, and I. Papamichail. Bi-Level
tions. Working paper. 2015. https://www.researchgate.net/publication/ Optimization in a Transport Network. Cybernetics and Information
279532065_Signal_control_module_in_CAPACITEL_for_isolated Technologies, Vol. 15, No. 5, 2015, pp. 37–49.
_intersections._Working_Paper. 27. Wolput, B., E. Christofa, and C. M. J. Tampère. A New Stochastic Meso-
22. Guidelines for Traffic Signals [Die Richtlinien fuer Lichtsignalanlagen model for Optimizing Multimodal Signal Settings for Networks. In Pro-
(RiLSA)]. German Road and Transportation Research Association, ceedings of the 2015 IEEE 18th International Conference on Intelligent
Cologne, 2003. Transportation Systems, IEEE, New York, 2015, pp. 1690–1695.
23. Liu, H., A. Skabardonis, W. Zhang, and M. Li. Optimal Detector Loca-
tion for Bus Signal Priority. In Transportation Research Record: Jour- The Standing Committee on Traffic Signal Systems peer-reviewed this paper.