You are on page 1of 17

KNOWLEDGE-BASED SYSTEM FOR DESIGN OF

SIGNALIZED INTERSECTIONS
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

By J. S. Linkenheld, 1 R. F. Benekohal, 2 and


J. H. Garrett Jr., 3 Member, ASCE

(Received by the Urban Transportation Division)

ABSTRACT: For an efficient traffic operation in intelligent highway systems, traffic


signals need to respond to the changes in roadway and traffic demand. The phasing
and timing of traffic signals requires the use of heuristic rules of thumb to determine
what phases are needed and how the green time should be assigned to them. Because
of the need for judgmental knowledge in solving this problem, this study has used
knowledge-based expert-system technology to develop a system for the phasing
and signal timing (PHAST) of an isolated intersection. PHAST takes intersection
geometry and traffic volume as input and generates appropriate phase plan, cycle
length, and green time for each phase. The phase plan and signal timing change
when intersection geometry or traffic demand changes. This paper describes the
intended system functionality, the system architecture, the knowledge used to phase
and time an intersection, the implementation of the system, and system verification.
PHAST's performance was validated using phase plans and timings of several
intersections, including the examples within the Highway Capacity Manual (HCM).
A comparison of delays and levels of service of the phase plans and signal timings
generated by PHAST to those of HCM indicate that very comparable optimal
solutions are being generated by the PHAST system. A potential use of the PHAST
system would be in the design of demand-responsive intersection controllers for
intelligent-vehicle highway systems (IVHS).

INTRODUCTION

Traffic congestion has become a major environmental problem in most


of America's largest cities. A properly designed signalized intersection can
help reduce traffic delay, driver discomfort, fuel consumption, and air and
noise pollution by efficiently using the capacity of existing streets. Inter-
section signal design involves two major tasks: (1) Selection of proper hard-
ware; and (2) operation of the signal system. The Highway Capacity Manual
(HCM) (1985), states, "The most critical aspect of any signal design is the
selection of an appropriate phase plan." The phasing and timing of traffic
signals are the two important factors affecting delay and level of service
(LOS). To create a good phase plan, one needs to: (1) Have experience in
signal design from which heuristic rules-of-thumb can be developed; (2)
understand the interaction between traffic flows; and (3) know how signal
systems operate.
Acknowledging these requirements, the H C M indicates "the establish-
ment of a phase plan is the most creative part of signal design." Because
of the need for judgmental knowledge in solving this problem, the appli-

'Project Engr., Triton Consulting Engrs., Ltd., Schaumburg, IL; formerly, Res.
Asst. Univ. of Illinois, Urbana, IL 61801.
2
Asst. Prof., Dept. of Civ. Engrg. Univ. of Illinois, 1201 Newmark Civ. Engrg.
Lab., 205 N. Mathews Ave., Urbana, IL.
3
Asst. Prof., Dept. of Civ. Engrg., Carnegie-Mellon Univ., Pittsburgh, PA 15213.
Note. Discussion open until August 1,1992. To extend the closing date one month,
a written request must be filed with the ASCE Manager of Journals. The manuscript
for this paper was submitted for review and possible publication on May 2, 1991.
This paper is part of the Journal of Transportation Engineering, Vol. 118, No. 2,
March/April, 1992. ©ASCE, ISSN 0733-947X/92/0002-0241/$1.00 + $.15 per page.
Paper No. 26588.
241

J. Transp. Eng., 1992, 118(2): 241-257


cation of expert-system technology is a viable means of automating this
process. In this study, knowledge-based expert-system (KBES) technology
is used to develop a system for phasing and signal timing (PHAST) of an
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

isolated intersection. Past expert systems addressing this problem focused


on proof of applicability of the KBES concept to intersection design prob-
lems and did not concentrate on knowledge acquisition. This project has
focused on acquisition and representation of the knowledge needed to solve
this problem more completely. In its present form, the system operates by
selecting the phase plan and timing characteristics as a function of the
demand on the intersection.
The steps that phasing and signal timing (PHAST) takes to generate the
phase plans and signal timings are similar to those of a traffic engineer. The
system considers the conflicting volumes and intersection geometry to decide
on appropriate phases. Once the complete phase plan is decided, the cycle
length and green times are allocated. The personal knowledge and expe-
rience of two of the writers, Linkenheld and Benekohal, were used in
developing the rules for phasing and timing. Additional knowledge from
the HCM was used in developing the rules for the timing of the intersection.
The HCM does not provide procedures for selecting phase plans.

BACKGROUND
Several prototype expert systems have been developed to demonstrate
how expert-system technology can be used in intersection design or analysis.
In addition, there has been previous work done in developing knowledge-
based expert systems for the task of design. PHAST has made use of con-
cepts developed in both sets of systems. A brief review of research in similar
areas is given in the following sections.

Intersection and Signal Design Systems


Bryson and Stone (1987) developed Intersection Advisor (IA), a proto-
type system that recommends geometric modifications for improving inter-
section operation. Chang and Tang (1989) developed INTEL for real-time
traffic control at isolated intersections. Chang (1988) developed a KBES to
suggest a suitable left-turn treatment among the permitted, protected, and
protected/permitted phases. Chang (1989) also developed a prototype sys-
tem to illustrate the application of KBES technology in the intersection
design process. Kirby and Montgomery (1987) described the need to identify
the rules and constraints for signal design in various countries, and the need
to incorporate them into a system. Mason et al. (1989) developed a prototype
system to check whether the signal warrants of the Manual or Uniform
Traffic Control Devices (MUTCD) (1986) are met. Radwan and Goul (1989)
developed a prototype system for Signal Control at Isolated Intersections
(SCII). Yeh et al. (1986) described the characteristics of a KBES and sug-
gested some potential application in the transportation field. Zozaya-Go-
rostiza and Hendrickson (1987) developed an experimental prototype KBES
for designing the operation of traffic signals at an isolated intersection
(TRALI). TRALI uses Webster's formula to compute cycle length, timing,
and delay. The authors tested TRALI's phase selection using different flow
rates, and some of the results were less than optimal. For instance, at a
four-way intersection it selected a protected left-turn phase for only one
approach, while all other movements were stopped. One would expect a
dual-left or a left-through phase in this situation. There are many HCM-
242

J. Transp. Eng., 1992, 118(2): 241-257


based software packages to compute delay and green-time allocation (Wal-
lace 1990). A limited number of these, e.g., SIGNAL 85 (Strong 1990),
attempt to decide a phase plan as well. However, these systems use algo-
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

rithmic procedures that select the best alternatives from a limited number
of choices.

Related Design Systems


Encore (Garrett and Jain 1988) is an object-oriented KBES for design
of power transformers that uses a heuristic search strategy (Newell and
Simon 1972) to perform redesign. The system starts by generating an initial
design and then applies modification heuristics to generate alternative de-
signs in response to insufficiencies of previously generated design alterna-
tives. Brown and Chandrasekaran (1986) developed a routine design en-
vironment, called DSPL, which supports four phases of design activities:
(1) Requirement phase; (2) rough design phase; (3) design phase; and (4)
redesign phase. The architecture of DSPL is quite similar to that within
PHAST. Mittal et al. (1986) have developed a design paradigm, called
PRIDE, that follows a generate-test-analyze-advise-modify strategy. The
knowledge base in PRIDE is organized around design goals. Dixon et al.
(1987) have developed an architecture for redesign, called Dominic, in which
the specifications of the artifact are given, after which an approximate or
rough design is developed.

PHASING AND SIGNAL TIMING (PHAST) KKNOWLEDGE-BASED SYSTEM

PHAST is a knowledge-based expert system that designs the phase plan,


cycle length, phase sequence, and green-time allocation for an isolated traffic
intersection. The sections that follow describe: (1) The functionality and
the user interface; (2) the architecture of and knowledge within PHAST;
(3) the implementation environments used; and (4) the verification of the
system.

System Functionality

Overview
The design strategy employed by PHAST is similar to that used by expert
traffic engineers: generate fairly good initial phase and timing plans, evaluate
the performance of the signalized intersection, then modify the plans to
address inadequate performance. Knowledge acquisition was done in several
discussion and interview sessions between the authors. From these sessions
came the heuristics leading to the rules for phasing and timing. Further
clarification and evaluation sessions were used to formulate the final rules.
PHAST takes the geometry, traffic volume, and other characteristics of
an intersection as input and generates (i.e., designs) a set of phases and
green times as output. During the input stage, the user is able to enter the
number of lanes per approach and the characteristics of each lane. After
describing the intersection characteristics, the user initiates the phase-design
process, which applies a collection of phasing and timing knowledge to the
described intersection. The phasing and timing knowledge is used to gen-
erate a set of phases, then assigns traffic volumes and green times to these
phases. After the phases and phase times have been assigned, an evaluation
of the service level for the phase intersection is performed to determine the
efficiency of the phase plan. If the service level can be improved, i.e.,
243

J. Transp. Eng., 1992, 118(2): 241-257


modifications are identified that improve the service level, then the phase
plane is modified and reevaluated.
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

User Interface
The user interface to PHAST consists of an input editor/interface and an
output delay. The input editor/interface consists of a graphic form-driven
input editor that allows the user to describe the intersection to the system
using a mouse, menus, and the keypad. The input editor/interface also
provides a short menu of system-level commands for loading files, solving,
and exiting the system. The output display presents the user with a graphical
interpretation of the phase-plan diagrams and cycle properties. These two
interfaces are further described in the following.
Input Editor/Interface. The input editor/interface, shown in Fig. 1, pre-
sents the user with a graphic depiction of a standard, four-legged intersec-
tion. All four approaches to the intersection can contain any number of
lanes; icons representing each lane in an approach (displayed as black boxes)
are created after filling the number-of-lanes data item for the approach (see
Fig. 1). These icons are used to ask for and edit lane information. For
example, Fig. 1 shows the lane editor being used to input and edit the
description of westbound lane 1. This editor is activated when the left mouse
button is depressed over the appropriate lane icon (westbound lane 1). To
create a T-intersection, the number of lanes may be left at the default value
of zero. The corresponding approach for each lane is automatically recorded.
The editor has slots for the information that is needed about each lane.
These slots include the lane type, volumes, width, and grade, and are further
explained in a later section. The area type of the intersection (central busi-

inpjUtlitt'in'doyf

FILEJWHE INTERSECTION-12

\
SOUTHBOUND nane N
NUMBER-OF-LANES
[D

HESTJOUNDJj

HESTB0UND-LANE1 HUHBER-0F-LANES | 2 |

TYPE | LEFT-THRU

LEFT-TURN-VOLUHE 150
THRU-VOLUHE 425
RIGHT-TURN-VOLUME

LANEJUDTH
0
12
1111
COHNANDS
HEAVY_VEHICLES 5
I LOAD I
GRADE

PARKINGJiOVEHENTS
0
0
NUHBER-0F-LANES
tn SOLVE
| SAVE
BUS_ST0PS 0
EDIT_COHPLETE| ] aurr

FIG. 1. PHAST Input Editor/Interface

244

J. Transp. Eng., 1992, 118(2): 241-257


ness district or not) is inherited by all of the lanes. All of these slots are
filled or edited by depressing the left button over the marker to be changed
and then typing in the new value.
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

Besides being able to input and edit lane and approach data, the input
editor/interface provides access to the rest of PHAST via a menu of four
button items within the input window (i.e., the area where the intersection
is drawn). These four button items are Load, Save, Solve, and Quit. When
the user depresses the left button over Load or Save, the appropriate action
is performed on the file indicated in the file name area of the input window.
When Quit is selected, PHAST is exited. When Solve is selected, all of the
lane and approach data input by the user is transformed into a form usable
by PHAST. PHAST then reasons with these data to generate a phase plan
and assign green times (descibed in a later section). When PHAST has
completed the design of the phase and timing plan, it displays these phases
graphically, as described in the next section.
Output Display. The output of PHAST is presented to the user in a
graphic format, as shown in Fig. 2, after the completion of the phase-plan
design, evaluation, and modification processes. The output consists of the
phase-plan diagrams and the properties of the cycle. The phase-plan dia-
grams are graphic drawings that illustrate the lane movements that are
allowed during each subphase. The corresponding green-time allocation for
each subphase is automatically displayed as well. Up to six subphases can
be displayed, and they appear in their proper sequential order. The cycle
length and lost time are items that the user would need to know and are
displayed through a separate window.

A- <^
>
^sT
green = 15 sec. green = 25 sec. green = 6 sec.

, ^
<
X- 4-

•f-
green = 6 sec. green = 30 sec.

cycle length = 91 sec.

lost tine = 6 sec.

FIG. 2. PHAST Interface Output Screen

245

J. Transp. Eng., 1992, 118(2): 241-257


System Architecture
The knowledge in PHAST consists primarily of sets of design rules for
generating phases and assigning green times to these phases. Hence, the
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

architecture of PHAST, shown in Fig. 3, is typical of most rule-based sys-


tems. It consists of: (1) Working memory, which contains the goals for
controlling the order in which the subproblems are addressed, a description
of the intersection, and the phase plan; (2) a knowledge base, which contains
control rules that govern the activation and deactivation of design goals,
phase design rules used to generate an initial intersection phase plan (phasing
and timing knowledge), evaluation procedures to find the intersection level
of service, and modification rules to adjust and improve the initial phase
plan; (3) an inference engine, which applies the knowledge in the rule base
to the current state of working memory in order to solve the signal design
problem; and (4) a user interface, which allows interactive, natural use of
the system by a traffic engineer. The following sections describe the first
two parts of PHAST in more detail; the user interface was described pre-
viously.

Working Memory
Working memory is used to store information about the problem being
solved and the state of the partial solution of that problem. The following
sections describe the various pieces of information found in working mem-
ory.
Design Goal Hierarchy. The process-control knowledge within working
memory consists of a hierarchy of goals (shown in Fig. 4) that control the
sequence of design steps taken by the system. A set of control rules operates
on the goal structure and determines when goals are to be activated and
deactivated. The design process starts when the user presents PHAST with
a description of an intersection to be phased and timed. The system gen-
erates an initial design by invoking a set of production rules that generate
a near-optimal phase plan (with respect to traffic volume). The system then
proceeds to find a reasonable cycle length and assigns green times. After
the initial signal design has been generated, the level of service of the
intersection is computed, using evaluation strategies specified in the HCM
(Highway 1985). If the performance of the intersection control plan is above

USER INTERFACE |

J f KNOWLEDGE BASE
WORKING MEMORY Design Process
Control Rules
Design Process Goal
Hierarchy
Initial Phasing and
Timing Rules
Intersection Description
Evaluation Rules
and Procedures
Intersection Control
Pian Description
Modification Rules

INFERENCE ENGINE

FIG. 3. PHAST System Architecture

246

J. Transp. Eng., 1992, 118(2): 241-257


Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

Resign dual left p h a s e ' } " " ^ < ^ _ d e S i g n left thru p h a s e d C j e s i g n red times^)

FIG. 4. Design Goal Hierarchy

Lane SB2
Lane SB1 intersection. approach-EB lane-EBl
Approach SB lane-EB2
approach-WB
lane-WBl
lane-WB2
Approach EB — Lane WB2 approach-NB
Lane EB1 lane-NBl
— Lane WB1 lane-NB2
Lane EB2 1 Approach WB lane-NB3

It
approach-SB lane-SBl
lane-SB2

Approach NB. Lane NB3


Lane NB2
Lane NB1

FIG. 5. Illustrative Example of Intersection Modeling Objects

the desired minimum level of performance, the system stops and presents
the design to the user. If the performance is below that specified, a set of
rules for modifying the control plan based on exhibited insufficiencies in
performance is invoked. When modifications are complete, the system stops
and presents the user with the best design it could generate.
Intersection Description. In PHAST, an intersection is described by a
collection of interrelated object-attribute-value (OAV) triplets. Fig. 5 shows
an illustrative example of the objects that would be used to represent an
intersection, which may contain up to four distinct approaches. Each ap-
proach has certain characteristics that are shared with, or derived from, its
component lanes. The OAV triplets represent the necessary links between
lanes and approaches. For example, an approach to the intersection has an
attribute named volume, which in turn has a value that is the actual volume
of the approach derived from the summation of the appropriate lane vol-
umes. Approach, lane, phase, and cycle are all represented as OAV triplets
in the PHAST system. A description of their attributes is presented in the
following sections.
247

J. Transp. Eng., 1992, 118(2): 241-257


Approach Object. The approach object contains information on the group
of lanes that enter the intersection from one direction. There is a unique
approach for each of up to four intersection legs. The approach is described
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

by a unique name, its relative position with respect to the other intersecting
approaches, the total volume of the approach, as well as the street volume
(sum of the east-west or north-south approaches). Each approach is asso-
ciated with a number of lanes.
Lane Object. The lane object completely describes the characteristics
of any single lane in an approach. Each lane is identified by a unique name.
Other attributes of the lane object are as follows: the type attribute, which
identifies the lane movement characteristics (left, through, right, or a com-
bination of these movements); and the approach attribute, which is a pointer
that links the lane to the approach of which it is a part. This latter attribute
allows all lanes to be identified in the approach system, regardless of the
number of lanes per approach. Each lane has attributes that list the phases
of which the lane is a part. Lanes may be included in more than one phase,
namely dual-left, left-through, and through phases. Right-turn lanes may
also be included in the protected left phases of the crossing street, and these
phases are also accounted for as attributes. Other attributes of the lane
object include the left, through, and right turn volumes, saturation flow for
the lane, and the volume/saturation flow ratio for the lane.The lane object
also has attributes that hold values necessary for computing the adjustment
factors to the saturation flow of each lane. These attributes are the grade
of the approach, the amount of truck traffic, the width of the lane, the
number of parking actions, the number of bus movements, and the area
location type (i.e., central business district or not). Each lane is placed in
the phase plan by referring by name to the phases that treat some portion
of its volume.
Intersection Phase-Plan Description.
Phase Objects. Like the intersection model, the phase plan is described
as a hierarchical collection of subphases that are eventually linked to lanes
as shown in Fig. 6. The phase plan is divided into two main phases, one for
each orientation (east-west or north-south). Each main phase has the fol-
lowing attributes: name, orientation, and the critical volume/saturation flow
ratio for all subphases in the main phase. Below the two main phases are
subphases. The subphase object, which refers back to the main phase object
of which it is a part, contains attributes about the phase times. Green, yellow,
and red times are all attributes of subphase. Other attributes are the phase
type (e.g., through, left-through, dual-left) and the volume/saturation ratio
of the subphase. The green time allocated to a phase is computed from
formula II.9-2 (see a later section) of the HCM (Highway 1985). The afore-
mentioned phases are assigned green time based on the appropriate cycle
length.

thru-phase
. east-west phase <^~- dual-left-phase
left-thru-phase
phase-plan\
thru-phase
north-south phase ^— dual-left-phase
left-thru-phase
FIG. 6. Phase-Plan Model

248

J. Transp. Eng., 1992, 118(2): 241-257


Cycle Object. The cycle length, an attribute of the cycle object, is com-
puted from a comparison of formula II.9-1 of the HCM (Highway 1985),
and a chart developed from heuristic relationships. The methodology of this
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

process is discussed in a later section. The cycle object also has the following
attributes from which the cycle length is computed: lost time, number of
phases, volume/saturation ratio, and critical volume/capacity ratio. The at-
tribute adjusted phase keeps track of the number of left phases that had
green time adjusted up to 6 s, so the cycle object can reapportion time to
the other phases.
The previously described objects are used to represent the intersection
and the intermediate states of the phase-plan design. The next section de-
scribes knowledge that creates subphases and assigns lanes and partial lane
volumes to those subphases, based on the initial intersection description
provided by the user.

Knowledge Base
Design Process Control Rules. The design process is controlled by a set
of control rules that activate and deactivate the goals shown in Fig. 4. The
control rules are grouped into two types: (1) Those that are not specific to
a goal and operate on the higher levels of the goal structure by making
subgoals active when a goal is active and deactivating a goal when all of its
subgoals have been completed; and (2) those that are goal-specific and
determine when a leaf-level goal (representing a goal requiring specific
action such as design or analysis) has been completed and subsequently
deactivates it. All domain rules refer to the goal to which they apply. Thus,
by activating and deactivating goals, the control rules determine which do-
main knowledge to consider in solving the problem.
Initial Phasing and Timing Rules. The first task of phasing and signal
timing is to take the intersection specifications from the user and derive
other further information from the user specs. Second, the intersection is
analyzed (using heuristics) to see if a signal phasing and timing plan is
needed, based on minimum volumes from the incoming approaches. If the
need for signal phasing and timing exists, the control rules will activate the
goal to generate the phases (e.g., a phase for the north-south approaches
and a phase for the east-west approaches) and the subphases (e.g., through
subphase for the north-south phase) for a given collection of approaches.
After the intersection has been phased, the control rules activate the goal
of establishing a cycle length and assigning green times to the generated
phases. These processes are described in more detail in the following sec-
tions.
Initial Computation Rules. The rules performing the initial computations
manipulate the data input by the user into information that the system can
use. These operations were established to reduce the amount of information
required by the system, and to avoid forcing the user to do preliminary
calculations. The operations performed by the initial computations are grouped
under three further subgoals:

1. Find left-turn difference—These rules compare opposing left-turn lanes


and compute the differences in their volumes.
2. Find approach volume—These rules sum all individual lane volumes to
obtain the volume of the approach.
3. Sum of approaches—These rules sum approach volumes to find the vol-
249

J. Transp. Eng., 1992, 118(2): 241-257


ume of the street flow (both directions). This street volume is used for justifying
the need for a signal.
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

Signal Justification Rules. The rules used in signal phasing and timing
justification were adapted from the MUTCD (1986) to establish minimum
thresholds that would justify the need for a signal phasing and timing design.
The rules do not constitute an actual warrant and are not intended to serve
that purpose. PHAST is used to design the phasing and timing of a signal
already established. The actual criterion that would warrant establishing a
signal should be performed outside of the system. The PHAST justification
eliminates the very low-volume intersections from analysis. The phasing and
timing justification should not be confused with signal warrant requirements.
Minimum volumes were established that appeared to be significant enough
to satisfy all of the warrants in terms of volume requirements. Street or
approach volumes above these minimum thresholds justify the need for
signal phasing and timing and activate PHAST. Volumes not meeting these
criteria will cause the system to end its computations and suggest to the
user that volumes may be too low. The intent of these rules was to prevent
unreasonably low street volumes to be input and run for a signal design
when none is needed.
Phasing Rales. The phase generation process, which follows the basic
structure of a phase plan as shown in Fig. 6, is basically a divide-and-conquer
process. The phase plan is divided into two main phases (north-south and
east-west), each of which is designed individually as if it were completely
unrelated from the other subphase. However, these two main phases are
related via the presence of conflicting right turns with left turns. Thus, after
each main phase is designed, any conflicts due to right turns are resolved.
For each main phase, the system determines if left-turn and through sub-
phases are needed and designs them if necessary. For the left-turn subphase,
two more subphases are possible: dual-left (where both left turns from
opposing approaches proceed simultaneously) and left-through (where the
left, through, and possibly right turning traffic from one approach proceed).
T-intersections and one-way streets receive special attention in the left-turn
design phase, which is described in the next section. Also receiving special
attention are exclusive right-turn lanes; issues involving right-turn interac-
tions are covered in a later section. This hierarchical treatment of phase
design permits the addition of further subdivisions of phases if deemed
necessary. Because the same rules are applied for each orientation (east-
west and north-south), only the knowledge used to develop the hierarchy
of subphases for one orientation is needed. More detailed descriptions of
the logic involved in phase design are presented in the following sections.
Left-Turn Phase Design Rules. The design of the left-turn phase is bro-
ken into two subtasks, the dual-left subphase design and the left-through
subphase design. The total volume of the left-turn lanes is the first criterion
that governs the dual-left phase selection. For the dual-left subphase selec-
tion, a phase is not given unless the volumes for two opposing left-turn
lanes are both above a certain minimum threshold. The current value is 75
vehicles per hour (vph), which was selected for the following reasons. For
an average cycle length of 90 s, there are 40 cycles per hour. The minimum
criteria for selecting the left turn is approximately two cars arriving per
cycle, or 80 vph. The value of 75 vph was chosen as our design criteria to
be slightly conservative. A second criterion governing this phase selection
is the cross-product of a left-turn lane and the largest opposing through

250

J. Transp. Eng., 1992, 118(2): 241-257


volume. This rule covers situations where left-turn volumes are not large
enough to justify a turn phase themselves but opposing through traffic is
very heavy. The rule fires when the cross-product exceeds 50,000. The dual-
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

left phase will not be chosen if there are no exclusive left lanes present.
The left-turn difference, which is the difference in volumes of the opposing
left-turn lanes, governs the selection of the left-through subphase. If the
left-turn difference is large (currently greater than 75 vph, and based on
the same reasoning as previously given), then a left-through phase will be
initiated to handle the excess flow. The left-through phase is also created
if there is an exclusive left-turn lane with significant flow, but no dual-left
phase was given. For example, a one-way street does not received a dual-
left phase since traffic flows in only one direction. The presence of the left-
through phase is determined by the left-turn difference, but through and
right-turn lanes can also be a part of the phase. Special rules were developed
concerning one-way streets and T-intersections. For these special cases, the
criteria for activation are not controlled by the left-turn difference, but rather
the maximum lane flow. There is no through phase given to these cases, so
the left-through phase must pass all of the traffic.
Through-Phase Design Rules. The creation of a through phase is broken
down into two further subgoals: (1) Justification of the through phase,
consisting of rules that prune away the through phase if the street has no
through volume, or is a one-way or T-intersection; and (2) addition of lanes
to the through phase, which assigns all appropriate lanes to the through
phase, including applicable left-turn lanes, through lanes, and right-turn
lanes for both directions of the street.
Phase-Interaction Rules. The phase-interaction rules address the issues
involving exclusive right-turn lanes. Exclusive right-turn lanes are able to
be included in some left-turn phases of the intersecting street, depending
on the lane location. If there is a right-turn lane that is in an approach to
the left of a left-turn lane, the right-lane movement creates no conflict with
the left-turn phase, and is included in the phase plan. Fig. 7 illustrates this
idea. We can assume that the dual-left phase, phase A, has been selected.
The right-turn movement, movement 2, does not interfere with the left-
turn movement, movement 1. Since there is no conflict, the right-turn lane
is included in the dual-left phase, but the phase is designated as a cross
dual-left phase to indicate the distinction that the right-turn lane is the
crossing orientation of the original phase.
Lane-Factor Adjustment Rules. The lane-factor subgoal invokes adjust-
ment rules to take the inputs of lane width, heavy vehicles, grade, parking

GEOMETRIC
CONFIGURATION

FIG. 7. Right-Turn Interactions


251

J. Transp. Eng., 1992, 118(2): 241-257


movements, bus stops, and area type to find the adjustment factors for the
saturation flow of each individual lane. The various lane factors all have an
effect on the volume of traffic that any one lane can handle. The ideal
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

volume, or saturation flow, of a lane is 1,800 vph. As the factors are applied,
the lane volume is adjusted upward or downward accordingly. The lane-
adjustment factor is based on various relationships derived from charts given
in the Highway Capacity Manual (1986) and finds the corrected saturation
flow of each lane.
Degree-of-Saturation Rules. The purpose of the degree-of-saturation rules
is to compare the critical volume/saturation (vis) flow ratio for the soon-to-
be-compared cycle. The subgoal begins by finding the vis ratios for each
lane is the system. It then analyzes the phase plan to look for the possibility
of overlapping subphases. Overlapping subphases occur when a left-through
subphase has been designed as part of the phase plan along with either a
dual-left or a through subphase. For the case of overlapping subphases, the
critical lane movements are analyzed and the largest sum of vis ratios for
conflicting movements becomes the vis ratio for the east-west or north-south
main phase. For the case of a nonoverlapping subphase, the phase vis ratio
is determined to be the largest vis ratio for all lanes present in the subphase.
The sum of the subphase v/s ratios becomes the v/s ratio for the east-west
or north-south main phase. The cycle v/s ratio is then computed from the
sum of the east-west and north-south main phase v/s ratios.
Cycle Length Rules. The assignment of green time to the subphases
begins with computations that determine the cycle length to be used in the
phase design process. First, the lost time for the cycle is computed by
assuming three s per discrete phase. Phases that overlap or that contain
lanes previously in another phase are considered nondiscrete. Total lost
time (L) is the sum of the lost times for all discrete phases. The cycle length
(C) is then computed using formula II.9-1 of the HCM (Highway 1985),
and is shown as (1).
LXc
C = (1)
Xc - 2

The critical volume capacity ratio (Xc) is assumed to be 0.85 for this
original computation, to simulate a moderately high traffic-volume condi-
tion. This formula has been found to be reasonable for midrange cycle
lengths, but as the summation of vis approaches the critical volume/capacity
ratio (Xc), and the denominator of the equation approaches zero, the cycle
lengths become unreasonable. To overcome this problem, a chart based on
a set of simple relationships was developed to provide a reasonable estimate
of the cycle length (see Fig. 8). For the relationships, the cycle length is
computed based on the actual volumes of the critical subphase flows. The-
cycle lengths computed by the formula and chart are then compared to each
other. If the difference is less than 20 s, it is assumed that the formula
method is in its valid range and that is used as the cycle length. This choice
keeps Xc, which was assumed at the start, at a reasonable level. If the
difference is greater than 20 s, it is assumed that the formula method has
yielded an unrealistic result, due to a very small value of the denominator,
so the chart result is used as the cycle length. If the chart value is used for
the cycle length, the assumption that Xc = 0.85 no longer holds, but since
the cycle length is now known, X, a variation of Xc, can be backcalculated.
252

J. Transp. Eng., 1992, 118(2): 241-257


Problem HCM PHAST
Intersection phase results p h a s e results
Example 1
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

C=70 sec. C=70 sec.


5
^jy?, , * 20
*
510 | J 7002r_
I V30 -*4
A>
#
^ o ] I 370 -* -2
3? ' 3 ^ 0 gl= 27 see. g2= 37 sec. gl= 22 sec. g2= 31 sec

Example 2 C= 50 sec. C= 60 sec.


100
<*N|f* •*~t^-»
700/^
*0
gl= 5.2 sec. el= 22 sec.
z:
-J
120
980 tit 785
2
*•

7v 40' '25 g2= 18.4 sec.g3= 20.4 sec g2= 6 sec. g3= 25 sec.
Example 3 C= 119 sec. C= 75 sec.
175 100
1
70*
840 «%20
4k 2 1 ^
gl= 6 sec. g2= 2 sec. gl= 9 sec. g2= 43 sec.

60J>

~ i 27C
1480
A 4 1= ^
g3= 71.7 sec.g4= 30.1 sec. e3= 17 sec.
90 v 120 80

FIG. 8. Cycle Length Determination

The value of X will be used in green-time allocation as an estimate of the


subphase volume/capacity ratio. This method may not produce the optimum
cycle length on the first try, but it does lead to a more reasonable value
than would be gained from the formula, under the condition where the
formula's denominator is very small.
Timing Rules. With the cycle length known, the timing for the phase
plan is able to proceed. Yellow time is assumed to be three seconds per
phase, which is generally considered to be the norm. Effective green time
is computed from formula II.9-2 of the HCM (Highway 1985), shown as
(2), with X (or Xc) for the cycle used to approximate the individual phase
vie ratio.
v C
(2)
s 'X
Green times for left-turn phases are to be six.s minimum to accommodate
at least two cars per phase. If a left phase is less than six s, it's time allotment
is increased to be a minimum six s and the remaining phases have their
253

J. Transp. Eng., 1992, 118(2): 241-257


green-time allotments redistributed accordingly. Effective red time is the
difference of the cycle length minus the effective green time.
Evaluation. Evaluation of the performance of the phase plan and timing
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

of the intersection is based on the delay and degree of saturation computed


using the design procedures in the Highway Capacity Manual (1985). Cur-
rently, a microcomputer version of the Highway Capacity Software (High-
way 1987) is used to perform this evaluation and the results are manually
fed back into PHAST's working memory. Eventually, it is planned to access
this software within the PHAST and perform the evaluation automatically.
Modification. Redesign occurs by applying a set of modification rules
that make adjustments to the control plan given that the desired level of
performance is not achieved by the previous design. Currently, no modi-
fication rules have been implemented. The system has only been developed
to the point where a high-quality initial design exists. However, as part of
continuing development of PHAST, modification heuristics are being col-
lected and codified into a body of knowledge that can be used to modify
the initial designs given patterns of intersection delay.

System Implementation
The PHAST expert system was implemented in OPS5 (Fergy 1981), a
LISP-based production system. OPS5 is a forward-chaining production sys-
tem, meaning its primary method of knowledge representation is the rule.
The rules in PHAST are divided into sets of rules for the different aspects
of phasing and timing as shown in Fig. 4 and described in a previous section.
The set of productions to be used are driven by a previously defined goal
structure. The system fires a set of production rules based on the current
active goal in working memory. When a rule is fired, the working memory
is updated (i.e., the phase plan is further designed), allowing another rule
to fire. The system terminates when all of the goals are satisfied.
The user interface was implemented using FRAMEKIT (Nyberg 1988),
a frame-based knowledge representation scheme written in LISP. Frames
were used to represent the various components of the user interface as well
as the intersection objects used to store the user input. The user interface
frames represented components of the interface, such as the input window,
the editor, editor-input objects, and active-region objects. To create the
PHAST interface, instances of these objects were created and given appro-
priate attribute values. Hence, by using frames to develop the PHAST user
interface, a too kit for creating any interactive graphic user interface for an
OPS5 program was developed. The editor objects were written so that they
referred to an editee—another frame object that stored information input
by, or edited by, the editor. These editees were frames representing the
approaches and lanes of an intersection. Each one of these editees was given
a method for generating OPS5 working memory elements from their frame-
based descriptions. Hence, when the user inputs data into the user interface,
he is actually building a set of FRAMEKIT frames corresponding to the
approaches and lanes of his intersection. When he saves or loads input, he
is saving these FRAMEKIT frames. Whenever he solves the intersection,
each of these intersection frames is sent a message to generate a corre-
sponding OPS5 working memory element (WME). Once these WME's have
been generated, the OPS5 command, run, is issued by the solve method
and OPS5 begins to execute the rule sets for phasing and green-time allo-
cation.

254

J. Transp. Eng., 1992, 118(2): 241-257


System Verification
The phase-selection portion of P H A S T has been tested by analyzing
several example intersections, including those in the Highway Capacity Man-
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

ual (1985). The traffic volumes, intersection geometry, green time, phase
plans, and cycle lengths of the H C M examples were used as inputs to the
PHAST system (shown in Fig. 9). P H A S T then provided an initial phase
plan, cycle length, and green-time allocation appropriate for the intersec-
tion. The PHAST designs were then used as inputs to the Highway Capacity
Software (Highway 1987) package for delay computations. The test results
have been very favorable. The H C M examples that contained solutions are
shown in Fig. 8, which compares the H C M phase selections with those of
PHAST. In each case, except for the third example, PHAST matched the
H C M phase selections. It happens that, for all cases tested so far, the
PHAST design provides for a comparable delay, although the phase plan
may be different. The test cases not shown here provided similar results.
More often than not, the cycle length has been chosen from the heuristic
relationships developed by the authors, rather than from the H C M meth-
odology.
The PHAST recommendations for effective green time, phasing, and
cycle length for each intersection were input to the Highway Capacity Soft-

cycle
length
(sec.)

500 1000 1500 2000


total volume on critical approaches, vph

FIG. 9. Test Cases

TABLE 1. Comparison of Delay and Level of Services Between HCM and PHAST
PHAST
Timing
Highway Capacity Manual Initial Design Modified3
Exam- Cycle vie Delay Level of Cycle vie Delay Level of Delay Level of
ple length ratio (s/veh) services length ratio (s/veh) services (s/veh) services
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
1 70 0.84 22.2 C 59 0.90 17.1 C 15.1 C
2 50 0.75 11.0 B 60 0.70 9.7 B not not
modi- modi-
fied fied
3 119 0.95 25.1 D 75 0.95 21.2 C 20.0 C
"These values were obtained after readjusting the green-time allocation according to
a simple heuristic procedure that simulates the modification process. They were computed
to emphasize the fact that the first values are only initial designs and are not meant to
be final. The cycle length was not changed.

255

J. Transp. Eng., 1992, 118(2): 241-257


ware (Highway 1987) to compute their level of service. Table 1 shows the
level of service analysis results comparing the level of service for the HCM
phases and a Highway Capacity Software analysis of PHAST's phasing and
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

timing results. Columns 6-9 in Table 1 list the results taken directly from
the solutions given by PHAST. Table 1 shows the vie ratios computed by
PHAST and HCM are similar. It is shown that the PHAST designs provide
different cycle lengths and smaller vehicle delays for all cases. The level of
service is comparable between the two methods. Columns 10 and 11 show
how better results can be obtained by applying the modification rules, which
were manually simulated in this case. It is important to note that the modified
timing is only an illustration to emphasize the point that the initial PHAST
design is not meant to be the final design and that better performance can
be achieved. Currently, modified timing results are mere reapportions of
green times. The green time is reduced from an approach with the lowest
amount of delay and is added to the approach with the highest delay. For
the future timing modification rules vie ratio and delay will be used. Mod-
ification rules will also be applied to cycle-length calculations, yielding fur-
ther improved performance.

CONCLUSIONS AND RECOMMENDATIONS

The signal design process consists of four major steps: (1) Phasing; (2)
timing; (3) evaluation; and (4) revision. The signal design problem possesses
a large number of viable solutions, but only a few result in a near-optimal
plan. Phasing requires the use of heuristic rules-of-thumb to determine what
phases are needed and how the volume should be assigned to them. A phase
plan that is far from optimal can lead to numerous iterations, and, quite
possibly, lead to a situation where the optimal phase plan can never be
reached. A knowledge-based expert system solves problems by using heu-
ristics to reduce the search over the problem's solution space and efficiently
find good solutions to the problem. Hence, expert-system technology can
be effectively used to address the signal design problem.
PHAST has been developed to solve the most creative part of signal
design (development of phase plan) using design heuristics to find a good
solution in a short period of time. In developing PHAST, the authors have
tried to exhaustively acquire and represent the knowledge needed to solve
the phasing and timing problem. In reaching this goal, PHAST uses methods
and techniques that would be familiar to the practicing traffic engineer,
including the philosophy of the HCM. The performance of PHAST to this
point has been favorable, although more testing needs to performed and
more rules need to be added.
A potential use of the PHAST system would be in the design of demand-
responsive intersection controllers, or applications with real-time intersec-
tion traffic control units. With an emphasis on the use of intelligent-vehicle
highway system and advanced traffic-management approaches, traffic en-
gineers will need smart tools for improved management of urban congestion,
particularly under varying demand conditions. The future step for the PHAST
system would be expanding it to a network of demand-responsive intersec-
tion controllers. Further testing and validation are required before field
implementation of the PHAST can occur. Also, work needs to be done to
enable PHAST to compute delay directly.
256

J. Transp. Eng., 1992, 118(2): 241-257


ACKNOWLEDGMENTS

The funding for this research effort has been provided by the Research
Downloaded from ascelibrary.org by "National Institute of Technology, Rourkela" on 10/29/23. Copyright ASCE. For personal use only; all rights reserved.

Board of the University of Illinois, Urbana, Illinois. The findings of this


research reflect the views of the writers only and do not necessarily reflect
the views of the University of Illinois. This interface was derived from a
system of window and region objects that were created by John Basten at
the University of Illinois for use as an object-oriented approach to con-
structing windowing environments.

APPENDIX. REFERENCES

Brown, D. C , and Chandrasekaran, B. (1986). "Knowledge and control for a me-


chanical design expert system." IEEE Trans. Comput., 19(7), 92-100.
Bryson, D. A., and Stone, J. R. (1987). "Intersection Advisor: An expert system
for intersection design." Transp. Res. Record, 1145, 48-53.
Chang, E. C.-P. (1988). "Deciding alternative left turn signal phases using expert
systems." 1TE /., 58(1), 21-27.
Chang, E. C.-P. (1989). "Interactive intersection design using expert systems ap-
proach." Transp. Res. Record, 1239, 10-16.
Chang, A. T. S., and Tang, R.-Y. (1989). "A prototype expert system for traffic
signal control in real-time." First Int. Conf. on Application of Advanced Tech-
nologies, C. Hendrickson and K. C. Sinha, eds., ASCE, 183-188.
Dixon, J. R., Howe, A., Cohen, P., and Simmons, M. K. (1987). "Dominic I:
Progress toward domain independence in design by iterative redesign." Engrg.
Comput., 2(3), 137-145.
Forgy, C. L. (1981). OPS5 User's Manual, Carnegie-Mellon Univ., Pittsburgh, PA.
Garrett, J. H., and Jain, A. (1988). "Encore—An object-oriented knowledge-based
system for trans-former design." Artif. Intell. Engrg., Des. Manuf., 2(2), 1-20.
Highway capacity manual (Special Report 209). (1985). Transp. Res. Board, Wash-
ington, D.C.
Highway capacity software user's manual. (1987). U.S. Dept. of Transp., Federal
Highway Admin., Washington, D.C.
Kirby, H. R., and Montgomery, F. O. (1987). "Towards a rule-based approach for
traffic signal design." Civ. Engrg. Syst., 4, Mar., 20-26.
Manual on uniform traffic control devices. (1988). U.S. Dept. of Transp., Washing-
ton, D.C.
Mason, J. M., Su, Y., and Kilareski, W. P. (1989). "Knowledge-based expert system
for traffic signal warrants." First Int. Conf. on Application of Advanced Technol-
ogies, C. Hendrickson and K. C. Sinha, eds., ASCE, 193-198.
Mittal, S., Dym, C. L., and Morjaria, M. (1986). PRIDE: An expert system for the
design of paper handling systems." IEEE Trans. Comput., 19(7), 102-114.
Newell, A., and Simon H. (1972). Human problem solving. Prentice-Hall, Engle-
wood Cliffs, N.J.
Nyberg, E. H. (1988). The FRAMEKIT user's guide, Version 2.0. 3rd Ed., Carnegie-
Mellon Univ., Pittsburgh, PA.
Radwan, A. E., and Goul, M. (1989). "Knowledge-based system development in
real-time traffic control." First Inter. Conf. on Application of Advanced Technol-
ogies, C. Hendrickson and K. C. Sinha, eds., ASCE, 189-192.
Strong, D. (1990). SIGNAL85 signalized intersection analysis and design. Strong
Concepts, Northbrook, IL.
Wallace, C. E. (1990). "The future of Highway Capacity Software, Part Two,
McTRans." 4(4).
Yeh, C.-L, Ritchie, S. G., and Schneider, J. B. (1986). "Potential applications of
knowledge-based expert systems in transportation planning and engineering." Transp.
Res. Record, 1076, 59-65.
Zozaya-Gorostiza, C , and Hendrickson, C. (1987). "Expert system for traffic signal
setting assistance."/. Transp. Engrg., 113(2).

257

J. Transp. Eng., 1992, 118(2): 241-257

You might also like