TRANSPORTATION RESEARCH RECORD I ]94

Comparative Analysis of Two Logics for

Adaptive Control of Isolated Intersections
FnNc-Bon LIN

Adaptive signal control has the potential to provide improved employed, the quality of the information used for making
control at isolated intersections. Adaptive control, however, timing decisions, the level of control efficiencies delivered by
has limitations due to its need to rely on estimated flow con- existing control devices, and the traffic flow patterns involved.
ditions for making signal timing decisions. Such estimated flow Field tests conducted so far have shown mixed but encour-
conditions always differ from the actual conditions, and the aging results for adaptive control. A test of Miller's algorithm
discrepancies can offset the benefit of having an elaborate deci- (8) at an intersection revealed that the resulting control effi-
sion making process in a control logic. Therefore, an issue can ciencies were poorer than those provided by vehicle-actuated
be raised as to whether it is necessary to rely on strenuous controls when the flows approaching the intersection were
decision-making processes for adaptive control. This study less than 1,300 vph. A test of modernized optimized vehicle
compares the relative merits of a simple queue based logic and
actuation strategy (1) for ten time-of-day periods resulted in
a logic that relies on a much more complicated procedure for
making timing decisions. It is found that the queue-based logic delay reductions of 5Vo to l2%o for seven periods, a delay
is nearly as effective as the more complicated logic. This finding increase of.7Vo for one period, and delay reductions of.20Vo
points to a direction for the development of new control logics and 30Vo, respectively, for the remaining two periods. The
-l that can be widely used to replace existing traffic'actuated implementation of the traffic optimization logic (4) at one
control logics. intersection yielded delay reductions of more than20Vo when
compared with a traffic-actuated operation.
Adaptive control, referred to herein, is a mode of control As demonstrated in these field tests, the incorporation of an
which relies on very short-term advance vehicle arrival infor- optimization capability into signal control does not necessarily
mation in an attempt to achieve real-time optimization of guarantee improved signal operations. The accuracy ofthe infor-
signal operations. Several adaptive control logics have been mation utilized to make timing decisions is also critical to adap-
tested in the field, implemented, or recommended for use at tive control. This can be a drawback, because adaptive control
isolated intersections. Examples of such logics include mod- usually relies on estimated flow conditions rather than on actual
ernized optimized vehicle actuation strategy (1), Miller's algo- flow conditions. To facilitate the estimation of flow conditions,
rithm (2), optimization policies for adaptive control (J), traffic it is necessary to place detectors several hundred feet upstream
optimization logic (4), and stepwise adjustment of signal tim- of the intersection in order to provide advance vehicle arrival
ing logic (5). The split, cycle, and offset optimization tech- information. Such detectors can provide perhaps no more than
nique (ó), which is intended mainly for signal coordination, 115 sec of advance information at mbst intersections. In order
has also been tested for the control of isolated intersections to overcome this limitation, some researchers (3, 8) have resorted
(n to the use of predicted vehicle arrival data to supplement the
It should be noted that, regardless of the level of sophis- detector data. This approach tends to introduce errors into the
tication of a control strategy, optimal signal operations can information that is used to make signal timing decisions. Even
never be achieved in a real life situation. The term optimi- if the estimation of the flow conditions is based entirely on
zation is often used casually to represent a process of searching detector data, the resulting estimates can be expected to deviate
for a better course of action. Such a process can be based on from the actual flow conditions. The discrepancies between the
a straightforward trade-off analysis in order to determine estimated and the actual conditions can be attributed to lane
whether the current green duration should be extended for a changes and to variations in vehicle speeds, queue discharge
short time interval (4). It can also be based on an elaborate headways, driver responses to signal change interval, and so
procedure to evaluate alternative signal switching sequences forth.
and, subsequently, to identify the best sequence for a rela- In light of this drawback, it is worth investigating whether
tively long (e.g., 100 sec) future period of time (3). This liberal strenuous decision-making processes can be replaced by sim-
interpretation of optimization is also adopted in this paper. ple decision rules for adaptive control. To address this issue,
two adaptive control logics are compared in this study. One
INTRODUCTION logicis stepwise adjustment of signal timing (SAST) (5), which
requires the evaluation of alternative signal switching sequences
The extent to which adaptive control can irnprove signal oper- in order to reach a signal timing decision. The other logic is
ations depends in part on the specific adaptive control logic based on the consideration of queue length and determines
whether the current green should be terminated by comparing
Civil Engineering Department, Clarkson University, Potsdam, N.Y. the expected maximum queue length of the current green
13676. phase with a threshold queue length.
 Lin

Time is T

Current green phase is i

exi s ts

Li )*u, 'I

i )ru, ' (1.)rì

and TL. > TL.

es ti mate
queue I engths
atT opt'imization

termìnate green
reached at
T+AT

extend green to
T+¿T, then enter
change i nterval

FIGURE I Decision-making process for stepwise adjustment of green.

SAST LOGIC either to extend the green beyond the current step or to ter-
minate the green at the end of that step. Referring to Figure
SAST logic relies on a binary choice decision-making process 2,let T be the beginning of a step. At least two types of data
for stepwise adjustment of signal timing. In this decision- are needed for making a timing decision. One type of data is
making process, time is divided into small intervals, or steps. the vehicle arrival sequence that is expected at the stop line
In each step, an analysis is made to determine whether the in several steps beyond Z. This type of data is derived from
current green should be terminated at the end of that step. vehicle arrival data obtained by the upstream detectors. The
The rationale for the development of this logic is discussed procedure for deriving such data is simple.
elsewhere (5). Let f, be the arrival time of a vehicle at an upstream detector
The decision-making process adopted in SAST logic for location and r the average travel time between the detector
stepwise adjustment of a green duration is shown in Figure and the stop line in the absence of interferences by signal
1. This process has four levels of decision-making activities, operations. Then, the expected arrival time of that vehicle at
which are marked in the figure as I, II, III, and IV, respec- the stop line is assumed to be A¡ : t¡ r. This expected arrival
tively. The first three levels employ simple decision rules which time can be used directly for decision making. It can also be
either permit the current green to be extended beyond the represented as one arrival in a specified step. The latter
first step or call for additional analyses. The data processing approach enables more efficient data processing. Therefore,
requirements for these three levels are very limited. Signal it is adopted in SAST logic. Following this approach, ,4, is
optimization, which is the last level of the decision-making transformed into one vehicle arrival in the ¿th step beyond
process, comes into play only if the first three levels fail to Tif ,4, falls in that step. Since the efficiency of adaptive control
choose a definitive course of action. can be sensitive to the errors in the vehicle arrival sequence
The manner in which SAST logic processes and utilizes that is used for signal optimization, the step size A,T should
information to reach signal timing decisions is described in be sufficiently small. Large step sizes will distort an arrival
detail below. sequence. Two second steps are a reasonable choice. In each
of such steps, the number of vehicle arrivals will rarely exceed
Data Acquisition and Processing one. On the other hand, the step size should be sufficiently
large in order to allow time for data processing, signal optim-
Time is divided into successive steps in SAST logic. Each step ization, and implementation of a timing decision.
is AZ sec in length. A decision must be made in each step Because the number of arrivals in each step is derived from
 TRANSPORTATION RESEARCH RECORD 1194

detector

FIGURE 2 Discretization of time for timing adjustment.

detector data, a finite amount of advãnce information is avail- the right-of-way before most of its queuing vehicles enter the
able at a given point in time. Referring again to Figure 2; if intersection. The result is poor signal operation. This problem
the average travel time between an upstream detector and can be eliminated when a threshold queue length is used to
the stop line is equivalentto M steps, then only those vehicles bypass signal optimization.
detected between T - MLT and Z will produce advance For example, if the threshold queue length is set at four
information for decision making at T. In other words, the vehicles, the queue lengths of the current phase can be reduced
amount of advance information available at T is MAT sec. to about four vehicles before other phases are allowed to
The second type of data needed for decision making is the compete for the right-of-way through signal optimization. This
expected queue length in each tratÏic lane at time Z. Such ensures that the queue lengths of every phase will not grow
queue lengths are determined from a traffic model, which is at excessive rates due to the existence of short green intervals.
an integral part of SAST logic, and are defined as the differ- A previous study (5) reveals that the best threshold queue
ences between expected cumulative arrivals and departures Iength to use appears to be about four vehicles. This implies
as measured at the stop line for a specified point in time. that it is best to allow the queue lengths of the current green
phase to be reduced to approximately four vehicles before
other phases are allowed to compete for the green.
Level I Decision Making To prevent exceedingly long green durations, SAST logic
also allows the imposition of a maximum allowable green G-u"
If competing demands for the right-of-way do not exist, there on the current phase. This G*"*, however, is imposed only
is no reason to terminate the current green phase. At a given when the maximum queue length of all completing phases
time 7, competing demands are considered to be nonexistent exceeds a specified threshold value.
if all the phases waiting for the right-of-way have the following
expected flow conditions:
Level III Decision Making
1.. There are no queuing vehicles in any lane at T, and
2. There are no vehicles expected to arrive at the stop liûe This level of decision making takes into consideration the
in n steps following l. queue lengths of the current green phase and those of all
competing phases. The current green in extended beyond Z
A reasonable value of ¿ is one such that nL,T is about 6 + AT if the following two conditions are satisfied:
sec. If a competing demand after a period of ¿AZ results in
a decision to terminate the green, nAT shorter than 6 sec may 1. The maximum queue length (L,)-"* of the current green
unnecessarily force an approaching vehicle to a complete stop phase I is longer than the maximum queue length (L.)*.. of
before it is given the right-of-way. all competing phases, and
2. The total number of queuing vehicles TL,of the current
green phase is larger than the total number of queuing vehicles
Level II Decision Making TL, of all competing phases.

This level of decision making is based on the maximum expected

queue length (1,)*"- of the current green phase aT T + LT. Level IV Decision Making
If this queue length exceeds a specified threshold value L,
the current green is automatically extended beyond T + LT, This level of decision making involves signal optimization.
subject to a maximum green constraint. This feature is impor- SAST logic allows minimization of the total delay either of
tant: if adaptive control relies exclusively on signal optimi- all vehicles or of the vehicles in certain critical lanes. The
zafion, a phase with a relatively low demand is likely to lose subject vehicles include the queuing vehicles at T and those
Lin
9

vehicles which a¡e expected to reach the stop line between Z this option are generated and evaluated one at a time. If the
and Z + MA,T. The critical lanes to be included for signal delay D, resulting from such a switching sequence is less than
optimization can vary from one step to another. They are or equal to D-,", it is more desirable to extend the green
determined according to the following criteria for each phase: beyond T + LT. In such a case, the current green is allowed
to continue unless the maximum green constraint prohibits
1. A lane that has a long queue length at time Z is more further extension of the green. If D, is greater than D_,n
critical than a lane that has a short queue. instead, another signal switching sequence is generated and
2. If two lanes have equal queue lengths at time f, the evaluated in the same manner until all alternative sequences
lane that has a larger number of expected arrivals between I associated with terminating the green at T + 2LT are exhausted.
and I + 2AZ is more critical. Following that, the option of terminating the green at T +
3. If two lanes have equal queue lengths at Z and equal nAT for n : 3, 4, . . . may be evaluated.
numbers of expected arrivals between T and T + ZAT, the SAST logic uses a decision variable N*"* to limit the max-
lane which is ahead in the data processing order is more imum number of options that are to be evaluated. For exam-
critical. ple, if N-"* : 3 is specified, only those signal switching
sequences involving the termination of the green at T + AT,
T + 2^T, and T + 3LT are considered for evaluation. With
Optimízation Process M steps of advance information, the value of N-,, can vary
from2to M.
The signal optimization process is illustrated in Figure 3. The
first task in this process is to examine the option of terrninating
the green af T + A?. This option leads to several alternative Generation of Switching Sequences
signal switching sequences. These sequences are generated
and evaluated in order to estimate the minimum delay D-,. Given that the current green is to be terminated at T * nAT
associated with this option. The next task is to determine (, : 1, 2, . . . , M), SAST logic does nor aftempr to generare
whether D-," can be reduced by extending the green beyond all feasible signal switching sequences for evaluation. Instead,
T + LT.This task is carried out byfirstconsidering the option it generates a small number of switching sequences that are
of terminating the green at the end of the second step, i.e., likely among the best few of all feasible sequences.
af T + 2A,7. The signal switching sequences associated with The process of generating signal switching sequence can be

option I

consider termìnating green at T+AT

generate alternative signal

switching sequences for option 'l

detennine minimum deìay Dr.,n

next option n=2

consider terminating green at T+nAT

generate a switching sequence

determine delay D

Dn <D mln ãlternatives

-
extend green
beyond T+AT
if Gmax is
not viol ated

next option
terminate green at T+AT

FIGURE 3 Signat optimization process.

10 TRANSPORTATION RESEARCH RECORD 1 194

the right-of-wav, the green is given to phase 2 which follows

the current green phase. The resulting signal switching sequence
phase I is likely to be inefficient, and this will be reflected in the
¿ timing decision.
A demand is recognized in SAST logic if any lane of the
(a) 0ptjon 1
phase in question satisfies either of the following conditions:

1. Expected queue length at the time the last change inter-

I
val is terminated (e.g., T + 3AT) is greater than zeto.
2
2. Expected number of arrivals within 4 sec after the ter-
2
minatisn of the change interval is greater than zero.
(b) 0ption ì
If phase 2 receives the green at T + 3AT, this green can
1 be extended by one step (Fig. 4a), or by two steps (Fig. 4b).
¿ Once the generated portion of a switching sequence reaches
the end of the last step, i.e., T + sLT, one of the following
actions is taken in generating the portion of the switching
(c) Opt ion 2
sequencesbeyond T + 5LT:

tlffi
^tlllll
r- 1.. If a signal change interval is in effect at the end of the
¡ |'.'. i I t--71 | I
last step, i.e., at + 5AT, that interval is allowed to be timed
:liiltl i out at or beyond T + sLT (Fig.4a,4d, and 4e). Afterwards,
(d) 0ption 3 the green is given to the next phase that still has vehicles
waiting to enter the intersection. This green is allowed to
1ltt1F-11¡-l= continue until all the vehicles in that phase are discharged.
2. lf a green interval is in effect at T + 5Af (Fig. 4b, 4c,
zllliii"-t i
and 4f), this green interval is extended beyond T + 5ATuntil
1 I t t I I I t.- all the vehicles are discharged.
(e) Option 4 3. The generation of a signal switching sequence is com-
pleted when all the vehicles included in the analysis are pre-
I sumably discharged. This point in time is denoted as P in
2 Figure 4.
3
Option 5

Estimation of Delays
FIGURE 4 Example of generated signal switching sequences.
The delay experienced by a vehicle is measured as the expected
departure time minus the expeçted arrival time at the stop
better described with the example shown in Figure 4. This line in the absence of interferences by signal operations. SAST
figure should be interpreted as follows: logic estimates only the delays of those vehicles which are
expected to reach the stop line by T + MLT. Therefore, it
is assumed that there are no additional vehicle arrivals beyond
1. The total number of signal phases is three.
2. Phase t has the green at time I. T + MAT.
3. The signal change interval following each green equals Delays are estimated simultaneously with the generation of
two steps. each signal switching sequence. When the front portions of
4. A bold ascending line indicates that a phase has the several signal switching sequences are identical, the delays
green, and a bold horizontal line signifies a signal change related to such portions are only estimated once in order to
interval. reduce the CPU time. For example, the first two signal switch-
5. Advance information is available for M : 5 steps beyond ing sequences depicted in Figure 4 have the same switching
pattern between T and T + 3AT. Therefore, the delays incurred
T.
6. Switching sequences marked as (a), (b), (c), (d), (e), in this period and estimated for the first switching sequence
and (f), respectively, are arranged according to the order in are used directly for the second sequence.
which they are generated. The total delay associated with a signal switching sequence
is the sum of the delays incurred in each step. The delays in
In Figure 4a, the current green is terminated at T + LT' each step can be estimated from the cumulative arrivals and
Subsequently, a signal change interval is timed out at T * departures both at the beginning and at the end of that step.
3AT. The green is then given to the next phase, i.e., phase To estimate such delays foreach lane, the following two quan-
2, lf that phase has a demand for the right-of-way. If phase tities are defined first:
2 has no demand, the green is given to phase 3 by skipping CA": Que(Î) (1)
phase 2, provided that phase 3 has a demand for the right-
of-way. If neither phase 2 nor phase 3 has a demand for CD.: g (2)
 Lin 11

where as the SAST logic. The only difference is that the optimization

CA" : cumulative number of expected arrivals at f,

process of the SAST logic (level IV) is replaced by a simple

Que(T) : queue length at Z, and

decision rule based on queue length. It should be noted that
the level II decision making of this queue-based logic can be
CD" = cumulative number of expected departures at
eliminated without affecting the resulting control efficiencies.
T.
This level of decision making, however, was not removed in
Next, the cumulative number of arrivals CA, at the end of this study when the relative merits of the SAST logic and the
Step I (i = I,2,3, . . .) can be determined as queue-based logic were evaluated.
The queue-based decision rule that replaces the optimiza-
CA'=CA,-1,+NAi (3)
tion process of SAST logic involves a comparison of the fol-
where Ná, : number of expected arrivals in step l. lowing two values:
For i : t,2,. . ., M, NAiis derived fromthe detector data.
For i > M, NAt is set equal to zelo because SAST logic does 1. (L,)^ * : expected maximum queue length of the cur-
not consider the delays of those vehicles not yet detected by rent green phase at I + AI, and
T. 2. Lr : predetermined threshold queue length.
The cumulative number of departures CD, at the end of
step I is determined as If the maximum queue length (L,)-", is less than or equal
to the threshold value L., the cr-rrrent green phase is termi-
cDi:cDi-1+NDt=CAi (4) nated at T + LT. Otherwise, the green phase is allowed to
where ND, : number of expected departures in step i. extend beyond T + LT, subject to a maximum green
In the original SAST logic, vehicle departures from the stop constraint.
line were treated as discrete events. In this study, the depar- In appearance, the queue-based control logic is as simple
tures of queuing vehicles are treated as a continuous variable as the gap seeking logic of traffic-actuated control. They dif-
in accordance with a nonlinear function of saturation flow. fer, however, in significant ways. Traffic-actuated control
After a queue is completely discharged from the stop line, attempts to terminate the current green when the arriving
the number of expected departures in a step is set equal to vehicles can no longer utilize the green efficiently. Unfortu-
the number of expected arrivals in the same step, i.e., nately, the gap-seeking logic of this mode of control can easily
ND,: NA'. misjudge the actual flow conditions. The queue-based logic
Equations 3 and 4 are applied in a stepwise manner, begin- also attempts to terminate the current green when the arriving
ning with the first step (i : 1) that starts at I. The combined vehicles can no longer utilize the green efficiently. However,
delay of the vehicles in a lane in step i is estimated as it requires the synthesis of detected vehicle arrivals to form
: a reasonably accurate picture of the conditions for decision
DELAY, (CA,_,. + CAt - CD,_, - CD)AT:T (5) making. Consequently, this decision-making process is more
intelligent than the gap-seeking logic of traffic-actuated con-
Choice of N^^* trol- For general application, the queue-based logic also needs
a more sophisticated vehicle detection system, because accu-
N^.* is used in SAST logic to limit the maximum number of rate estimation of queue lengths is not necessarily a simple
options to be evaluated in the signal optimization process problem.
(Fig. 3). If N-"* : 4 is chosen, the signal optimization is The idea of using queue length as a control criterion is not
limited to the evaluation of the options of terminating the new (9, l0). A conventional wisdom of queue-based control
current green at T + LT, T +,zLT, and Z + 4AZ, respec- is to extend the current green until the vehicles in the gov-
tively. For the example depicted in Figure 4, this means the erning queue are completely discharged into the intersection.
optimization process will be forced to terminate after the first In analyzing SAST logic, it was found that allowing queuing
five switching sequences are generated and evaluated. Of vehicles to dissipate completely before calling for optimization
course, the optimization process may end as soon as the third can be detrimental to the control efficiency (5). Therefore, it
switching sequence is generated and evaluated. is necessary to choose a proper threshold queue length for
Given that M steps of advance information are available, implementing the queue-based logic.
SAST logic allows the evaluation of up to M options of ter-
minating the current green. It is not necessary, however, to
consider all the options. Simulation analyses (5) reveal that the ASSESSMENT OF CONTROL LOGICS
efficiencies of SAST-based signal operations are not very sen-
sitive to the choice of N-*. With detectors installed at a distance A microscopic, event-oriented simulation model was used to
of 400 ft upstream of the intersection, N-* : 2 is sufficient to compare the performances of SASTlogic and the queue-based
achieve a high level of confrol efficiencies. With a longer detec- logic. A major component of this model is a flow processor,
tor setback of 600 ft, a larger N-u* may be desirable. Never- which generates vehicle arrivals and processes vehicles down-
theless, N** of 3 or 4 is sufficient in such a case. stream according to prevailing signal indications. Another
major component is a signal processor, which allows various
signal logics to be implemented for evaluation. Delays esti-
A QUEUE-BASED CONTROL LOGIC mated from the model agree very well with the estimates
obtained from Webster's formula (11) for pretimed operations
The queue-based logic aîalyzed in this study is depicted in (5). Simulated delays under traffic-actuated operations have
Figure 5. This logic retains the same decision-making structure also been compared with delays measured on six intersection
.,''i
t2 TRANSPORTATION RESEARCH RECORD T194
' |:
.ì Time is T

Current green phase is i

competì ng.

(1.)
' l'max > L

(Li )*u*
' (1.)*u,
and TL. ' TL.

es ti mate
queue 1 engths impose G*u,
atT
(1.)
'r'max- > L"
T

¡ = l+aT
reached at
T+^T

extend green to
T+AT, then enter
change i nterval

FIGURE 5 Queue-based control logic'

approaches (5). They differ by less than 3Vo when actual ft upstream of the intersection to detect vehicle arrivals. The
vehicle arrival sequences are used as inputs into the simulation step size ATwas seL at2 sec and N*u* was limited to 2. The
model. The differences between the simulated and the mea- maximum green was set at 60 sec. This constraint, however,
sured delays can be greater if vehicle arrivals are generated took effect only when a queuing vehicle was stopped by a red
from specified flow rates. light. In addition, the current green was extended automat-
To provide an insight into the desirability of replacing traffic- ically if the maximum queue length of the current green phase
actuated rrintrol with adaptive control, SAST logic was com- exceeds four vehicles.
pared with ^onventional loop occupancy control logic. This Referring to Figure 6, it can be seen that the advantages
comparison' based on various hourly flow patterns which of the SAST based control over the loop occupancy control
were subjected to either two- to four-phase control. The num- can vary from one hourly flow pattern to another. When the
ber of lanes associated with a phase was varied from two to flow rates are low, and the delays under the loop occupancy
four. The lane flows ranged from 100 to 750 vph per lane. control are less than 10 sec/veh for two-phase operations, the
The total flows approaching the intersection were in the range SAST-based control can be only as efficient as the loop occu-
of 600 to 5,600 vph. No conflicting movements were present. pancy control. For four-phase operations, the SAST-based
The four-phase control had two protected left-turn phases to control cannot be expected to deliver significant improve-
accommodate vehicles in continuous left-turn lanes. The sim- ments when the delay under the loop occupancy control is
ulated vehicle had an average approach speed of 40 ft/sec. less than 20 sec/veh. Under moderate to heavy flow condi-
For the loop occupancy control, the maximum allowable tions, however, the SAST-based control can improve the con-
green was set at 60 sec. When two-phase operations were trol efficiency in some cases by more than 207o; the mosf
encountered, 50-ft detectors were used if a phase was asso- likely level of improvement appears to be in the range of 8Vo
ciated with four lanes, and 70-ft detectors were used if a phase to L5Vo. Since traffic-actuated signals are not necessarily uti-
was associated with two lanes. For four-phase operations, 50- lized to their best ability, the actual improvement through
ft detectors were used in left-turn lanes, while 7O-detectors adaptive control can be greater than what is implied in
were used in others. The extension interval was set at zelo Figure 6.
seconds for all the cases examined. These timing settings and Traffic-actuated control based on loop occupancy or vol-
detector configurations yield near optimal operations under ume density logic can be very efficient under light flow con-
heavy flow conditions. ditions. Under heavier flow conditions, two problems may
For SAST-based operations, 5-ft detectors were placed 400 emerge. One problem is the failure of the control to allow
 Lin 13

o 4-phase .4-phase
50
oZ
.2-phase
c
o
o
o .: "s5;; 940
o40
;
; Õ
o tr
ã o
o oJU
30 L
o c
! o
c O
o !
a o
ã20 Æ¿o
o
5
o
J
O

lu zo 30 40 50 60

Loop Occupancy Control Deìay, sec/veh

l0 20 30 40 50
SAST Control Deìay, sec/veh
FIGURE 6Delays of SAST control versus delays of loop
FIGURE 7 Delays of queue-based control versus delays of
occupancy control.
SAST control (detector setback = 400 ft).

most queuing vehicles to enter the intersection due to pre- for all the hourly patterns tested in this study. At the present
mature termination of the green. This problem can arise when time, however, it is unknown which combination of N-"* and
short detector lengths, short vehicle intervals, or short max- L will result in the best overall operation for an intersection
imum greens are employed. On the other hand, excessively with a wide range of flow conditions. Despite this limitation,
long greens may resuh because of the actuation of detectors it should be noted that, with the exception of a few tested
by vehicles not in a queue. Such vehicles cannot utilize the flow patterns that have low flow rates, the queue-based con-
intersection capacity as efficiently as queuing vehicles. This trol consistently performs better than the loop occupancy con-
problem can become rather acute when many lanes are asso- trol. This characteristic can be exploited for adaptive control
ciated with a signal phase. of intersections where short auxiliary lanes, opposed left turns,
Adaptive control can alleviate these weaknesses of traffic- or frequent right-turn-on-red maneuvers exist. At such inter-
actuated control through real-time optimization. Similarly, sections, reliable advance information cannot be obtained for
the queue-based logic, as shown in Figure 5, can also prohibit all approach lanes. In these cases, real-time information on
vehicles that cannot efficiently utilize the intersection capacity queuing flows may be obtained and used to complement advance
from extending the green. To facilitate a comparative analysis information in order to produce efficient signal operations.
of SAST logic and the queue-based logic, the same simulation
model was used to determine the best threshold queue length
L, that should be used for the level IV decision-making of CONCLUSIONS
the queue-based logic. The best threshold queue length was
found to be in the range of 0.5 to 1.5 vehicles. Adaptive control has the potential to improve the existing
Based on a threshold value ofL, : 1.5 vehicles, the delays level of signal control efficiencies at isolated intersections.
produced respectively by SAST logic and the queue-based Generally, adaptive control requires a logic to identify flow
Iogic for a number of hourly flow patterns were estimated conditions and to use the identified conditions for making
through simulation. The results are shown in Figure 7, where intelligent timing decisions. This process of control usually
it can be seen that, for hourly flow patterns that have delays results in the use of estimated flow conditions for making
under 20 sec per vehicle, the SAST-based control is slightly signal timing decisions. Estimated flow conditions always deviate
better than the queue-based control. Under heavier flow con- from actual conditions. The detrimental effects of flawed
ditions, however, the queue-based control can sometimes per- information on signal timing decisions cannot be compensated
form better than the SAST-based control. for by the use of a strenuous process of searching for better
The ability of the queue-based logic to deliver reasonably signal operations. Therefore, it is pertinent to examine whether
high control efficiencies is not without a logical basis. Never- simple decision rules can be effectively used to replace a more
theless, one reason that the queue-based control can some- strenuous decision-making process for adaptive control.
times deliver better signal operations than the SAST-based In comparison with conventional loop occupancy control
control can be found in the use of N-"* : 2 and a threshold under simulated conditions, the SAST logic, which uses advance
queue length of L : 4 vehicles for the SAST-based control. information in a vigorous process for making signal timing
This combination of l{-"* and L is not necessarily the best decisions, can provide significantly better signal operations.
L4 TRANSPORTATION RESEARCH RECORD T 194

The level of improvement varies with a number of factors, for Isolated Intersections. Trffic Engineering and Control,
Yol. 27, No. 7/8, 1986, pp. 385-387.
but it tends to be higher when heavier flows are encountered.
2. A. J. Miller. A Computer Control System for Traffic Network.
Under the same simulated conditions, the simpler queue-based Proc., 2nd International Symposíum on Theory of Road Traffic
logic can produce comparable results. This implies that real- Flow, London, pp. 201-220.
time information on queuing flow can be used to produce 3. N. H. Gartner. OPAC: A Demand-Responsive Strategy for Traffic
improved signal operations. This understanding is important Signal Control . In Transportation Research Record 906, 'IRB'
Nãtional Research Council, Washington, D.C., 1983' pp.75-81.
in the development of a versatile adaptive control logic. 4. K.-L. Bang. Optimal Control of Isolated Traffic Signals. Zrafic
With the possible exception of the modernized optimized Engineering and Control, July 1976, pp.288-292.
vehicle actuated strategy (1), none of the adaptive control 5. F. B. Lin, N. Wang, and S. Vijayakumar. Development of an
strategies mentioned above can be effectively utilized for the Intelligent Adaptive Signal Control Logic. Presented at the Engi-
neering Foundãtion Conference on Management and Control of
control of intersections where short turning bays, opposed left
Urban Traffic, Henniker, N.H., June 74-19,1987.
turns, or right-turn-on-red maneuvers exist. At such inter- 6. P. B. Hunt, D. I. Robertson, R. D. Bretherton, and M. C. Royle.
sections, reliable advance information cannot be obtained for The SCOOT On-Line-Traffic Signal Optimization Technique.
all traffic movements. Under the circumstances, it would be Trffic Engineeríng and Control, Yol. 23, No. 4' 1982,
logical to use real time information on queuing flow, as well pp. 190-192.
7. Þ. Carden and M. McDonald. The Application of SCOOT Con-
as other advance information, for decision making. A major
trol to an Isolated Intersection. Trffic Engineering and Control,
challenge to facilitate such a use of information is to develop Vol.2, No.6, 1985, pP. 304-310.
a vehicle detector system that can provide reliable real-time 8. L. de la Breteque and R. Jezequel' Adaptive Control at an Iso-
and advance information at all or most intersections con- lated Intersectiôn-A Comparative Study of Some Algorithms.
Traffic Engineering and Control, Vol. 20, No. 7, 1979' pp. 361-
trolled currently with traffic-actuated signals.
363.
9. Caltrans. Diamond Interchange Program, Software Documen-
tation. Implementation Package, FHWA, U.S. Department of
ACKNOWLEDGMENT Transportation, IP-80-5. 1980.
10. D. O.-Gerlough and F. A. Wagner. NCHRP Report32: Improved
Criteria for Trffic Signals at Individual Intersections. HRB,
This study is sponsored in part by the National Science Foun- National Research Council, Washington, D'C., 1967.
dation and in part by the UPS Foundation. 11. F. V. Webster. Traffic Signal Settings. Road Research Technical
Paper No. 39. Road Research Laboratory, 1958.

REFERENCES

1. R. A. Vincent and C. P. Young. Self-Optimizing Traffic Signal Publication of this paper sponsored by Committee on Traffic Flow
Control Using Microprocessors-the TRRL'MOVA' Strategy Theory and Charccteristics.