Fuzzy Sets and Systems 116 (2000) 11–22

Signal control using fuzzy logic

Jarkko Niittymaki ∗ , Matti Pursula
Helsinki University of Technology, Transportation Engineering, P.O. Box 2100, FIN-02015 Hut, Finland

Received December 1998

Abstract
Applications with fuzzy logic being used in controlling trac signals have been designed already since the 1970s. The
strength of fuzzy logic lies in its capability of simulating the decision-making process of a human, a process that is often
dicult to de ne with traditional mathematical methods. The results of FUSICO-project have indicated that the fuzzy trac
signal control can be the potential control method for signalized intersections. c 2000 Elsevier Science B.V. All rights
reserved.

Keywords: Trac signals; Fuzzy logic; Multivalue logic

1. Introduction • to validate the fuzzy control principles and to cal-

ibrate the membership functions of the linguistic
Trac signal control is one of the oldest application variables using simulation and eld trials, and
areas of fuzzy sets in transportation engineering. The • to develop the principles of the fuzzy adaptive
published applications are mainly theoretical, but there signal controller.
is active research going on in this area [2,3,5,6,8 – 10]. The aim of this paper is to discuss the fuzzy signal
The subject of our FUzzy SIgnal COntrol (FU- control in general, to present the rst results of our
SICO) project at the Helsinki University of Tech- research project, and to show the preliminary general
nology, Laboratory of Transportation Engineering, is rule base and the structure of control algorithms for
the application of fuzzy control to trac signals at the complicated intersections, especially for the signal-
individual intersection level. The main goals of the group control.
research are
• to make a general theoretical analysis of fuzzy traf-
c signal control
2. Framework of fuzzy signal control
• to formulate generalized fuzzy control rules for
signal-group control using fuzzy logic
2.1. Application for trac signal control

∗Corresponding author. Tel.: +358-9-4513809; fax: +358-9- The theory of fuzzy sets is based on graded con-
4515019. cepts to handle uncertainties and imprecision in a par-
E-mail address: jarkko.niittymaki@hut. (J. Niittymaki). ticular domain of knowledge. The graded concepts are

0165-0114/00/$ - see front matter c 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 5 - 0 1 1 4 ( 9 9 ) 0 0 0 3 4 - 2
12 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

useful since real situations are not very often crisp and to red or remain as a passive green. The passive green
deterministic, and they cannot be described precisely. can be terminated by con ict signal groups.
Fuzzy logic control allows linguistic and inexact Fuzzy signal group control in our case works in the
trac data to be manipulated as a useful tool in de- same way as the traditional control, but the extensions
signing signal timings. The fuzziness of signal control are adjusted by a fuzzy extender, and the phase se-
can be divided into three levels: input, control, and quences are selected by a fuzzy selector. The main
output level. In the input level, we have only a partial principles of fuzzy control with the main principles of
picture of the prevailing trac situation through mea- phase ring control are compared in Fig. 2.
surements. In the control level, we have many pos- In general, the fuzzy rules are working at three
sibilities, and we do not know which one of them is levels:
exactly the right or the best one, because we cannot
explain the cause–consequence relationship of signal
2.2.1. Trac situation level
control. In the output level, the correct control crite-
The trac situation is divided into three di erent
ria are not known (for example, extension gap). In
categories: oversaturated, normal and low demand.
general, fuzzy control is found to be superior in com-
The de nition of trac situation is done using upper
plex problems with multi-objective decisions. In traf-
level fuzzy rules:
c signal control, several trac ows compete for the
same time and space, and di erent priorities are often
if VOL is any and min(OCC) is high
set to di erent trac ows or vehicle groups.
then TS is oversaturated or
The base principle of fuzzy signal control is to
if VOL is low and max(OCC) is zero
model the control based on human expert knowledge,
then TS is low or
rather than modeling of the process itself. The design
if VOL is any and max(OCC) is normal
of a fuzzy controller for this system requires the ex-
then TS is normal,
pert knowledge and experience of trac control in
formulating the linguistic protocol, which generates
where VOL is the trac volume of the last 5 min, vph;
the control input, to be applied to the trac control
OCC is the occupancy of the rst detector during the
system.
last 5 min, %; and TS is the trac situation.
The control policy of the di erent trac situations
2.2. Signal group control
is di erent. The control policy of oversaturation is the
capacity maximizing, and the control policy of low
In Finland, Germany, The Netherlands and in other
trac situations is First In First Out (FIFO).
Nordic Countries, the principle of signal group control
is used in trac signal control. Signal group control
is more exible than the traditional stage control and 2.2.2. Phase and sequence level (fuzzy phase
therefore more adaptive to various trac conditions. selector)
The signal group control does not have exactly de ned The main goal of this level is to maximize the ca-
stages, it has only primary phase pictures (Fig. 1). pacity by minimizing intergreen times. The basic prin-
The traditional vehicle-actuated control of isolated ciple is that “signal group can be kept in green while
intersections attempts continuously to adjust green no disadvantages to other ows occur”. This is also
times. Detectors give the input-data. The basic tim- called “the method to use extra green”. The main deci-
ing parameters at each phase in traditional vehicle- sion of this level will be the right termination moment
actuated control are minimum green, passage time of the green.
interval (extension interval) and maximum green. The decision moment is the moment, when the
In the base situation, the green signal group gives at green of the rst signal group of phase A can be ter-
least the minimum green time. If the demand is suf- minated, so that the rst signal group of phase B=C=D
cient, the green time can be extended stepwise with can be started. Secondly, the decision will be checked
the lengths of extension interval to the maximum when the last signal group of phase A is ready to
green. After green extension, the signal group can go terminate.
J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22 13

Fig. 1. Signal group control vs. phase control.

Fig. 2. Main principles of fuzzy control and phase ring control.

14 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

The second goal of this level is to determine the

right phase order. The basic principle is that the phase
can be skipped if there is no request or if the weight
(W (p)) of this phase is low. This means that if the
normal phase order is A-B-C-A the fuzzy phasing can
for example give the orders A-B-A-C-A or A-C-A-B-
C. The rules are a little bit complicated, when there
are four phases, but the principle is the same as in
the rules of three phases. The general principles of the
rules are
• if W (p) is very high then phase p will be the next
one,
• if W (pi ) is high and W (pj ) is zero then (i) will be
the next one,
• the maximum waiting time of each vehicle cannot
be too long,
• otherwise the order will be as planned.
The rules of four phase control are
if W (B) is high
then next phase is B or
if W (B) is medium and W (C) or W (D) is oversat
then next phase is W(higher)
if W (B) is medium and W (C)+W (D) is very high Fig. 3. The three levels of fuzzy signal control.
then next phase is C or
if W (B) is low and W (C) or W (D) is high
then next phase is W (higher) or
if W (B) is low and W (C) + W (D) is mt(medium)
then next phase is C or (mt = more than)
if W (B) is zero
then next phase is C.
The fuzzy factor W (X ) is the fuzzy factor for re-
quired green time/phase. The exact speci cation will
be done later. The rst suggestion could be the sum
of the queued vehicles’ waiting time. Basically the
second and the third signal group are working to-
gether. The green extension need is discussed in the
third level (Fig. 3).
Fig. 4. Decision tables for multi-phase control.
2.2.3. Green ending level or extension level (fuzzy
green extender)
The main goal of fuzzy rules of this level is to if S is low
determine the rst moment to terminate a signal group.
(1) and W (red) is low and GRN is short and GAP
The basic idea is not to terminate during the queue
is long then extend probably
discharging. This means that each vehicle has to stop
···
only once at each intersection. The fuzzy rules are
(27) and W (red) is big and GRN is long and GAP
if S is high then extend certainly or is small then terminate certainly,
 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22 15

3. The ÿrst applications of FUSICO-project

3.1. Signalized pedestrian crossing

The essence of fuzzy logic lies in its ability to han-

dle linguistic information by representing it as a fuzzy
set. As such, it is known to be suited for emulating
the control process normally done by a human being.
The developed algorithm of the signalized pedestrian
crossing attempts to emulate the decision process of an
experienced crossing guard. The crossing guard is al-
ways faced with two con icting objectives: minimum
delays to the pedestrians and minimum delays to the
vehicles while keeping maximum safety for both par-
ties. As the number of pedestrians and the number of
vehicles increase, the decision becomes increasingly
dicult. Fuzzy logic control res many soft rules at
the same time and takes the decision, which is in the
compromised position.
Based on the observations of Niittymaki and
Kikuchi [6], the proposed fuzzy control algorithm
appears to work well. The results show that the fuzzy
controller provides the pedestrian friendly control
while keeping vehicle delay smaller than the conven-
Fig. 5. Fuzzy inference of multi-phase control. tional controls. In other words, the fuzzy controller
provides a compromise between the two opposing
objectives, minimum pedestrian delay and minimum
vehicle delay, in accordance with the proportion of
where W (red) is the weight of red signal groups, GRN the pedestrian and vehicular volumes. The theory of
is the running time of the green signal group, GAP is signalized pedestrian crossing, the fuzzy algorithm,
the last observed gap between two approaching vehi- the membership functions, and the results are given
cles according to detection at 100 m before the stop in the paper of Niittymaki and Kikuchi [6].
line and S is the indicator of the discharging queue The conventional controls that are tested in the
(minimum discharging gap at the stop line). paper represent two opposite cases: The pedestrian
The basic idea is that the consequences have some friendly case (so-called “fast pedestrian signal”,
kind of uncertainties. Therefore the nal consequence which gives only one extension interval after the
of each rule could be terminate certainly (0.00), ter- signal group is ready to terminate) and the vehicle
minate probably (0.25), either or (0.50), extend prob- friendly case (traditional gap-seeking method). The
ably (0.75) and extend certainly (1.00). The decision fuzzy control, on the other hand, represents a com-
tables of multi-phase control are shown in Fig. 4. promise of these two cases by providing friendliness
After rule modi cation, the rules are located in these to both clients (pedestrians and vehicles) depending
ve groups, and the maximum value of membership on the trac volume. This nding is consistent with
function of each group will be selected. The COG- the characteristics of most other applications of fuzzy
method (center of gravity) calculates our extension control; the fuzzy control looks for a compromise in
ratio, which will be compared with the extension cri- the multi-objective problem environment. Further, it
teria (for instance 0.50). The extension criteria can be is interesting to note that such a compromised control
di erent for each control policy. The fuzzy inference is executed without speci c limits imposed on the
of multi-phase control is shown in Fig. 5. rules of the fuzzy control.
16 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

Another aspect that is worth examining when There are only two input variables for the fuzzy rule
comparing fuzzy control with conventional control base:
is robustness and adaptivity. The conventional signal A = number of approaching vehicles at the mo-
control requires setting of a large number of para- ment t (veh), or vehicles between detectors
meters, like minimum and maximum times of each D01 and D10 of green phase approaches
signal group and the logic of detectors. In the case of Q = number of queuing vehicles at the moment t,
the fuzzy logic controller, the number of parameters or stopped vehicles of red phase approaches.
can be reduced, and the meaning of each parameter The fuzzy rules of our application are
can be realized easily. This is possible due to the
use of the membership function that covers a range After minimum green (5 s)
and, as a result, the conclusions of the rules over- if A is zero then terminate immediately
lap. This fact makes more than one rule to re for a or if A is a few and Q is a few
given input, and the outcome is derived as a com- then EXT is short (3 s)
promise of the conclusions of more than one or if A is mt(a few) and Q is any
rule. then EXT is medium (6 s)
or if A is many and Q is any
3.2. Two-phase vehicle control with pedestrian then EXT is long (9 s).
signal-group
After the rst extension (ext1 + min gr)
The general rule at the signalized intersection is “the if A is zero then terminate immediately
fewer phases, the better”. Two are as few as you can or if A is a few and Q is a few
have and still have a meaningful trac signal installa- then EXT is short (3 s)
tion. It turns out that two phases is the ideal number. or if A is medium and Q is any
It is still true that there are more two-phase intersec- then EXT is medium (6 s)
tions than any other type. This is because signals with or if A is many and Q is any
two phases do the basic job of assigning the right- then EXT is long (9 s).
of-way and leave the motorist and pedestrian on their
own from there [7]. After the second extension (ext1 + ext2 + min gr)
The results of Pursula and Niittymaki [9] showed if A is zero then terminate immediately
that the fuzzy control algorithm can be competitive or if A is a few and Q is a few
against traditional vehicle actuation with the exten- then EXT is short (3 s)
sion principle. The compared fuzzy algorithm was or if A is medium and Q is lt(medium)
the algorithm of Pappis and Mamdani since 1977. then EXT is medium (6 s)
These results without any calibration of member- or if A is many and if Q is lt(medium)
ship functions were fairly good and showed that then EXT is long (9 s).
this kind of fuzzy control is suitable for many other
applications. After the third extension (ext1 +ext2 +ext3 +min gr)
The aim of our study is to compare our FUSICO if A is zero then terminate immediately
with the “best-known vehicle actuation algorithm”. or if A is mt(a few) and Q is a few
The traditional algorithm has two detectors 140 and then EXT is short (3 s)
60 m before stop line. The algorithm works using the or if A is medium and Q is lt(medium)
“gap seeking” method. In our fuzzy application, two then EXT is medium (6 s)
detectors are located per each approach lane. The lo- or if A is many and Q is lt(a few)
cation of the rst one is 100 m (D10) and the sec- then EXT is long (9 s).
ond one is at the stop line (D01). This means that
After the fourth extension (ext1 + ext2 + ext3 +
we know how many vehicles are approaching the stop
ext4 + min gr)
line within next 6–8 s. The minimum time of signal
if A is zero then terminate immediately
group is 5 s.
 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22 17

or if A is mt(a few) and Q is a few

then EXT is short (3 s)
or if A is medium and Q is lt(a few)
then EXT is medium (6 s)
or if A is many and Q is lt(a few)
then EXT is long (9 s).

The basic principles of these rules are to adjust

cycle time and to divide the cycle into the green
times of phases, and to emulate the performance of an
experienced policeman.
Fig. 6. Main objectives of trac signal control and their e ects
3.3. Multi-objective decision making on other goals.

Fuzzy control has proven to be successful in prob-

lems for which an exact mathematical modeling is There are many ways to determine the cost function
hard or impossible but an experienced human oper- of trac signals
ator can control the process. The trac signal con-
trol is a problem of this category. In particular, fuzzy C = C(time=delay) + C(environment=fuel)
logic control is suited for problems, which have many + C(safety=accidents):
and often con icting objectives with information (in-
put) given in linguistic terms. This trait is suited for In our project, the aim is to use fuzzy logic in decision
dealing with a control problem at a complex intersec- making. Basically, there are two ways to do that:
tion with many approaches and vehicle movement re- (1) A rule base, where the input-parameters take
quirements. Therefore, the result in this paper should into account the goals of the control algorithm.
provide an impetus to develop fuzzy control logic for An example of this is the control algorithm of
more complex intersections. a fuzzy signalized pedestrian crossing given in
There are three main objectives of signal control: this paper.
maximize safety, minimize delays and minimize envi- or
ronmental disadvantages. The problem is that the op- (2) A rule base, where the nal decision will
timum of each objective is achieved in di erent cycle be done using the fuzzy methods among the
times [1]. The problems of signal control can be de- multi-objectives. The example is that the -
scribed as a triangle, where the main objectives are nal decision will be done using the fuzzy
on the corners. Each main goal a ects other goals as methods between three objectives; trac
shown in Fig. 6. uency (delay), trac safety and environ-
The gure means that if the minimizing of delays mental e ects. Good trac uency means
is the main goal, the e ects to other goals are little that the average delay is small and the length
negative (−). The only positive e ect is between the of maximum queue is not long. Good trac
environment and safety (+). In other words, the en- safety means that the percentage of stops is
vironmentally e ective trac signal control can also not high and the risk of rear-end collision
be safe, because the cycle times of the environmen- is small and the pedestrian waiting time is
tally e ective trac signals are quite long. The long not long. Good environmental e ects can be
average cycle time means that the number of amber achieved, if the percentage of stops is small
intervals is smaller, and the risk of rear-end collisions and the number of zero delay vehicles is high.
is smaller. The biggest problem is that the environ- The nal decision can be made using, for
mental or safe control strategy does not give a good example, the min–max method among these
delay result (–). The average delay can be even 40% three multi-objectives. The defuzzi cation
bigger than the optimum delay. should be planned carefully. It is not even
18 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

important that all objectives are acting at the simulation of adaptive signal control is done by con-
same time. For example, during the rush hour necting a real controller to the microcomputer (PC)
the trac uency and the trac safety can based simulation system. For the development of new
be more important than the environmental control methods, an internal controller system has
aspects. been included into HUTSIM. This system, called as
We have these three objectives in our application, HUTSIG, works so that the controller object has
and each objective has its own rule base. The rst de- some measurement functions that are used to collect
tector of each approach is located suciently far away and analyze incoming detector data. The calculated
(approx. 160 m) from the stop line. The early detec- indicators of the trac situation are then transmitted
tion and the knowledge of vehicle kinematics make it to the control logic for timing decisions which are put
possible to model the trac situation during the next into force by the group oriented signal control [4].
9 s (maximum extension is 9 s in each interval). Com-
bining these three di erent rule bases is the determi- 4.2. Theoretical background for simulations
nation of extension made. The number of intervals can
be determined using the estimation of the maximum The fuzzi cation process involves the scale map-
green time, N = max:green=9. For instance, the nal ping of the measured input variables into the cor-
membership function values of each rule base for each responding universes of discourse. The preliminary
extension are as shown in Table 1. membership functions for two-phase vehicle control
The extension of this multi-objective case is 5 s are shown in Fig. 7. The functions were formulated
(maximum) in this interval. The extension according by trial and error. The basic ideas of the theory of
to delay rules would have been 5 s, according to envi- fuzzy sets are used for the quantitative interpretation
ronmental rules 6 s, and according to safety rules 4 s. of these instructions as well as for the decision mak-
This kind of multi-objective decision procedure with ing process. The operators mt(more than) and lt(less
fuzzy logic can be used both in the extension level and than) are de ned as follows. If A is a fuzzy set de ned
the phasing level. The rst results of the fuzzy multi- on the real line
objective signal control have been very promising. The
systematical testing and development of the rule base R1 = {xi };
will continue. In this case, we need more experience
A (xi ), is its grade of membership function and x0 is
to de ne a correct defuzzi cation method, because the
the element of R1 for which A(xi ) is maximum, then
centroid methods (like COG) are also available.
lt(A) and mt(A) are fuzzy sets de ned as

0; for xi ¿x0 ;
4. Simulation study of two-phase vehicle control lt(A) (xi ) =
1 − A(xi ) for xi ¡x0 ;

4.1. Simulation in trac signal control evaluation and


0; for xi 6x0 ;
The testing of a new control scheme such as the mt(A) (xi ) =
1 − A(xi ) for xi ¿x0 :
problem at hand requires not only the algorithm for
control but also a microscopic simulation model, This means that lt(A) or mt(A) = not(A), and lt(A)
which allows testing of many control schemes un- and mt(A) = 0.
der a realistic setting. In this respect, a sophisticated Since the output of rules is a set of membership
simulation model is indispensable for development functions associated with the output sets, the de-
and testing of an advanced signal control algorithm. fuzzi cation strategy is required in order to obtain
Our simulation package HUTSIM gives versatile a crisp control action. There are several commonly
possibilities to test di erent trac signal control al- used defuzzi cation strategies, even though there is
gorithms with each other. HUTSIM is a simulation no systematic procedure for selecting a defuzzi -
package that has been developed at the Helsinki Uni- cation strategy. The defuzzi cation method of our
versity of Technology. In the original design, the two-stage control algorithm is the center of gravity
 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22 19

Table 1

Extension (s)
0 1 2 3 4 5 6 7 8 9

Delay 0.0 0.3 0.5 0.7 0.9 0.9 0.8 0.6 0.4 0.3
Environ. 1.0 1.0 1.0 1.0 1.0 1.0 1.0 0.8 0.8 0.8
Safety 0.0 1.0 1.0 1.0 1.0 0.9 0.8 0.8 0.8 0.8
Result (min) 0.0 0.3 0.5 0.7 0.9 0.9 0.8 0.6 0.4 0.3

Fig. 7. Preliminary membership functions of two-stage control.

method (COG) because we are looking for some The results showed that the extension principle is
kind of compromise for the di erent extensions. The quite a good trac signal control mode in the area of
centroid–defuzzi cation method nds the output’s very low trac volumes. The results also indicated
center of mass (= compromise). that the application area of fuzzy control is wide. This
result is the same as the result indicated in the paper
of Pursula and Niittymaki [9].
4.3. Results of the simulations In 1996, we compared the traditional extension
principle with the fuzzy control algorithm of Pappis–
We compared the eciency of our FUSICO with the Mamdani [8]. The biggest di erence between the
traditional vehicle-actuated control (called extension results of our control algorithm and the control algo-
principle), since this algorithm is as realistic as possi- rithm results of Pappis–Mamdani is that the percent-
ble, and we do not know of any more e ective control age of stops of our control algorithm is smaller than
algorithms at the moment. Fig. 8 gives the average the percentage of stops of the extension principle.
delay and percentage of stops of the intersection with The delays are also slightly smaller (−10 –20% in the
two equal approaches as a function of trac volume tested area 100 –1200 vph) in our case than the delays
(trac volumes N–S and W–E are approximately the of the Pappis–Mamdani algorithm.
same). The simulation time of each case was 7200 s
and the simulated intersection was an isolated inter-
section of two one-way streets with 2 + 2 lanes. The 5. Conclusions and discussion
speed distribution of approaching vehicles was a Gaus-
sian with mean value about 40 km=h. The simulated Our FUSICO-project continues. The aim is to
trac was exactly the same in both the cases. move step by step to more complicated trac signal
20 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

Fig. 8. Comparison of the FUSICO-algorithm and the traditional extension principle.

systems. The general rule base for trac signal are not reported. One important step in the evaluation
control will develop with time, and also the fuzzy project of fuzzy control will be the eld test at a real
controller will become better with time. So far, the intersection. This test will be done at the signalized
published applications of fuzzy control have been intersection of Helsinki City.
mainly theoretical. The fuzzy control has been tested The results between Pappis–Mamdani fuzzy con-
sometimes in the real intersections, but the test results trol algorithm and traditional vehicle-actuated control
 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22 21

indicated that fuzzy control is suitable for many other mise of the results of these two conventional controls.
control algorithms. The long cycle times of the tra- One basic advantage of fuzzy control is that it res
ditional extension principle indicated the weakness many soft rules simultaneously and makes a decision,
of this kind of gap-seeking control. However, the re- which o ers the compromise. The control algorithm
sults of stopped vehicles of Pappis–Mamdani control with three di erent main goals (minimize delay, max-
were higher than the results of traditional vehicle- imize safety and minimize environmental e ects) has
actuated control. Based on these experiences, we already been developed as a part of the next phase.
developed our own control algorithm for two-phase The testing of this algorithm is in progress.
vehicle control. The main goals of this control were The calibration of membership functions is one im-
to adjust the cycle time and to divide the cycle into portant sub-goal of our project. Normally, the prelim-
the green times of phases. According to these results, inary calibration of the membership functions is done
the extension principle is good in the area of very low by trial and error. One goal of the future is to calibrate
trac volumes, but the results indicated that the ap- used membership functions using the neural networks
plication area of fuzzy control is very wide. The dif- or some other relevant systematic method. Neural net-
ference between our FUSICO-control algorithm and works have recently been recognized as an important
Pappis–Mamdani control algorithm is that our con- tool for constructing membership functions, perform-
trol algorithm gives smaller number of stops than the ing operations on membership functions, constructing
extension principle or Pappis–Mamdani control. Ba- fuzzy inference rules, and other context dependent en-
sically, this means that the FUSICO-algorithm should tities in fuzzy set theory. In general, constructions by
also have smaller fuel consumption and better trac neural networks are based on learning patterns from
safety than the traditional vehicle-actuated control sample data. The data for the calibration comes from
algorithms or Pappis–Mamdani control algorithm. simulations.
The comparison between the FUSICO-algorithm and The important step in the evaluation project of fuzzy
the SOS-algorithm (fuzzy logic against mathematical trac signal control will be a eld test at a real inter-
optimization) could be very interesting in the near section. This test will be done at the signalized inter-
future because at the moment both the algorithms section of Helsinki City. This step is very important
have proved to be better than traditional control because a successful test means that the fuzzy con-
algorithms. troller is ready to work in reality.
The results of this paper and our past work have
indicated that fuzzy signal control can be the poten-
tial control method for isolated intersections. Nakat- References
suyama et al. [5] and Chiu [2] have shown that fuzzy
control will be competitive at coordinated signal sys- [1] R. Akcelik, Fuel eciency and other objectives in trac
tems and networks. We believe that fuzzy methods system management, Trac Engng. Control 2 (1981) 54 – 65.
are well suited to almost all kinds of signal control, [2] S. Chiu, Adaptive trac signal control using fuzzy logic,
Proceedings of the Intelligent Vehicles Symposium, 1992,
and the biggest bene ts can be achieved in more com-
Detroit, pp. 98–107.
plicated intersections and environments than our ap- [3] S. Kim, Applications of Petri Networks and Fuzzy Logic
plications at the moment. The speci c plans for more to Advanced Trac Management Systems, Polytechnic
complicated isolated intersections are already avail- University, New York, 1994, p. 141.
able, and the rst plans for aerial fuzzy signal con- [4] I. Kosonen, HUTSIM – Simulation Tool for Trac Signal
Control, Helsinki University of Technology, Transportation
trol have been done. Theoretically, fuzzy control has Engineering, Publication 89, Espoo, 1996, p. 121.
also proved to be superior in complex problems with [5] M. Nakatsuyama, H. Nagahashi, N. Nishizuka, Fuzzy logic
multi-objective decisions. Trac signal control is a phase controller for trac junctions in the one-way arterial
typical process, where trac ows compete for the road, Proceedings of the IFAC Ninth Triennal World
same time and space, and di erent objectives can be Congress, Pergamon Press, Oxford, 1984, pp. 2865–2870.
[6] J. Niittymaki, S. Kikuchi, Application of fuzzy logic to
reached in di erent trac situations. The results of the control of a pedestrian crossing signal, Transportation
the fuzzy signalized pedestrian crossing have proved Research Record No. 1651, Transportation Research Board,
that fuzzy control provides timing, which is a compro- Washington, DC, 1998, pp. 30–38.
22 J. Niittymaki, M. Pursula / Fuzzy Sets and Systems 116 (2000) 11–22

[7] F.L. Orcutt, The Trac Signal Book, Prentice-Hall, in: The Fourth Meeting of the EURO Working Group on
Englewood Cli s, NJ, p. 158. Transportation, University of Newcastle, Newcastle upon
[8] C. Pappis, E. Mamdani, A fuzzy logic controller for a trac Tyne, UK, 9–11 September, 1996, p. 14.
junction, IEEE Trans. Systems Man Cybernet. SMC-7 (10) [10] T. Sayers, M.G.H. Bell, T. Mieden, F. Busch,
707–717. Verkehrsabhangige Lichtsignalsteuerung mit Fuzzy-Logik –
[9] M. Pursula, J. Niittymaki, Evaluation of trac signal control ein modulares, praxisorientiertes Verfahren. Paper
with simulation – a comparison of the Pappis–Mamdani fuzzy vero entlicht auf der HEUREKA, 1996, p. 11.
control vs. vehicle actuation with the extension principle,