306 THE 1987 INTERNATIONAL CONFERENCE: OF “THE SYSTEM DYNAMICS SOCITY. CHINA
A PROTOTYPE EXPERT SYSTEM FOR SYSTEM DYNAMICS MODELLING
Mitsuo KISHI, Keitaro NODA, Yuji YAMASHITA, and Katashi TAGUCHI
Department of Naval Architecture, College of Engineering,
University of Osaka Prefecture, Sakai, Osaka 591, JAPAN
ABSTRACT
In this study, a prototype expert system to support system dynamics
modelling is designed by organizing ‘the knowledge structures of : generic
patternized expectations and the tuleS‘on:how to construct system dynamics
models. The system is a production-rule-orierited consultation system written
in PROLOG. The proposed. system covers the system conceptualization in part,
system modelling, and. generation of simulation program, Brief executing
processes of the proposed system are:) 1)° Extracting concepts(nodes) within a
system by perceiving action/decision ing and by inferring the causal
relations(links). 2) Preparing a .causal-loop diagram of the system
automatically by interconnecting the! causal’; vrelations and by eliminating
inappropriate links. 3) Transforming the causal—-ioop diagram into a flow-
diagram automatically, and generating a Similation program. The proposed
system has a knowledge base of facts acquired in the systems modelling, to
facilitate the modelling of a system related to the ones dealt with in the
past. Some application examples are provided to vérify the applicability of
the proposed system.
1. INTRODUCTION
When analyzing -a multi-correlated problem related to technological,
social and/or | ecological . items,’. it’; may be appropriate to employ not. a
conventional - extrapolation model, but. a. structural model, e.g. system
dynamics .modél... (Forrester 1961). :In:system:dynamics, it is. possible to
simulate ~-system behaviour quantitatively using DYNAMO (Pugh 1976) after the
system modelling (systém ‘identification)...:However, “system. modelling is...
presented by specialists basing on their. own knowledge and experience ‘about
both .the system and-system dynamics, that is to say not.a systematic ways
Therefore, some heuristic .computer’aided:-techniques for structural modelling,
such ‘as ISM, Cognitive Map. (Axelrod. 1976, Kishi .1986), etc., have. been
proposed, but they are not so precisé ‘for. system dynamics modelling.
In’ this study, a “prototype expert system to support system dynamics
modelling is designed ..by organizing the knowledge .structures of generic
patternized expéctations and.the-rules.on how to construct system . dynamics
models. The system is a production-rule-oriented consultation system encoded
in PROLOG (Clocksin 1983). PROLOG; which is a programming language based on
predicate.. logic (Chang, .1973), is becoming popular for Artificial
Intelligence research. -And -it is-good for processing causal: relations in
system dynamics models (Nilsson’1980, Nolan 1986, Elzas 1986).
The process of system ‘dynamics inquiry consists of two phases as shown in
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 307
[PROBLEM DEFINITION ¢—__—_
CONCEPTUAL L
Coverage of the os = sisi CONCEPTUALIZATION ¢—————4
Proposed System #38
-
MODEL REPRESENTATION ————~ — REFINEMENT
7
MQDEL BEHAVIOR
ECHNICAL
we EVALUAT LON eed
| __ POLICY ANALYSIS AND MODEL USE
Figure 1 Process of system dynamics inquiry
(Andersen 1980,Roberts 1983)
Figure 1 (Andersen 1980, Roberts 1983): i) Conceptual phase, which addresses
the problem definition and system conceptualization. ii) Technical phase,
which addresses the system modelling, simulation of system behaviour,
evaluation of model validity, and system analysis. The conceptual phase is
critically important; however, it is close to being an art. For the above
process, the proposed system in this study covers the system
conceptualization in part and the model representation.
Brief executing processes of the proposed system are: 1) Extracting
concepts(nodes) within a system by perceiving action/decision making and by
inferring the causal relations(links). 2) Preparing a causal-loop diagram of
the system automatically by integrating the causal relations and by
eliminating inappropriate links .so as to be precise for system dynamics
models. 3) Transforming the causal-loop diagram into a flow-diagram by
identifying system levels, rates, auxiliaries and parameters automatically.
4) Generating a simulation program (in BASIC) semiautomatically by defining
equations on the causal relations and specifying initial values for
variables, etc.
To facilitate future modelling work about a system related to the ones
dealt. with in the past, the proposed system has a knowledge base of facts
acquired in the systems modelling. . The facts in knowledge base are the
causal relations in systems and the mathematical equations defined on
variables.
An operation test is given to examine the validity and applicability of
the proposed system. Moreover, a system dynamics model for maritime
industries is presented.
2, “SYSTEM IDENTIFICATION
The “first phase in system dynamics modelling is the problem definition
shich defines a model purpose, system boundary, and level of concepts
elements of a system) aggregation (Andersen 1980, Forrester 1980, Starr
980, Roberts 1983). This phase is critically important; however, it is close
o being an art. In the following, the subsequent phases after the problem
efinition are investigated.
308 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
2.1. Extracting Causal Relations
After the problem definition, there lies the phase of system
conceptualization which involves, for example, listing the concepts of a
system, extracting the causal relations between them, and identifying the
feedback structures in conjunction with time delays. "System dynamics deals
with change" (Forrester 1980). Any change is induced by some action
(including phenomena). Therefore, in this study, the system conceptualization
is pursued by perceiving action/decision making and by inferring the causal
relations between them. The concepts extracted from a system are expected to
be measurable or potentially measurable.
Parameter
r——~ State node Level
L—— Auxiliary
Node
Action node Rate
Figure 2\-Classification of nodes
Information 1 oun | ; — State 1
Information 2 t g Action State 2
: a t ot
~ (Physical Taws ) Jaws
Figure 3 . Process of action making
Intentional link
r—— Action making link
Action result Vink
L_. Simple reiation. link
Link eo Non-intentional link
Figure 4 Classification of Tinks
In the system conceptualization phase, a causal-loop diagram is to be
prepared. The concept of a system and the causal relation between concepts
correspond to the node and the link in the diagram, respectively. The nodes
_ (concepts) can be classified into "action" or "state" node. Figure 2 shows
the relationship among the nodes and the variables in system dynamics models,
The process of action making is expressed schematically in Figure 3, Action
occurs owing to some information/input, and the action changes the state of
the object. The action making. process is considered as a fundamental unit of
causal relations in systems. The links between nodes can be classified into,
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 309
so to call it, "action making", "action resulting", or "simple relation (i.e.
irrelevant to action making)" link, and further, the action making links are
classified into "intentional" or "non-intentional" link (see Figure 4). The
whole structure of a system is depicted by selecting the action making units
and by interconnecting them directly or through the simple links among them.
The process for selecting action making units is, for example, as follows: i)
extract the action making unit with explicit decision maker, ii) do that with
explicit action maker, iii) do that without explicit decision and action
makers (i.e. phenomenon).
It can be said that there is a great variety of data structures to
represent causal relations (Nolan 1986). In this paper, the data structures
of action, state, and information nodes are given in the form of predicate
logic form (Chang 1973) as follows: ,
Action(AN, OBJ, AM, DM, SYS) qa
State¢SN, SBJ) (2)
inform(IN) (3)
where AN is the name of an action,
OBJ the object of the accion, AM/DM
the action/decision maker of the
action, SYS the (sub-)system which
contains the action. SN is the name
of a state, SBJ the subject
(substance) of the state, and IN
the name of information. "AN" and
"SYS" are employed as keywords to
retrieve the information about
causal relations from a knowledge
base (library/catalogue) of dynamic
structures of systems (see 2.3.).
The data structure to represent a
link between nodes is:
Extract State nodes,
9, = inode j|State node}
‘xtract Information nodes,
93 = (node k| Information node}
Link(INN, TNN, LK, SIGN) (4) ia; = (node m[the node (Os) without
an inputting node n(«95)
where INN is the name of the such as Link(nm))
initial node, NN the name of the
terminal node, LK the kind of the
link (see Figure 4), and SIGN the
sign of the effect of INN on TNN
(positive or negative). It matters
little if there are null inputs to
AM, DM, SYS in Eq. (1), LK, SIGN in
Eq. (2).
In the next place, extraction
of causal relations in the upper
Reccgnize system boundary nodes
from the nodes <Q,
Q, = {node n|System boundary node}
xtract Information nodes about
the nodes « 9% nity,
Qq = (node pjthe Information node
extracted here
stream of the information nodes is ~ 4
pursued. The algorithm for
extracting the causal relations is
shown in Figure 5. Figure 5 Algorithm for extracting
After the extraction of causal causal relations
relations, all the information
310 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
nodes are classified into action or state nodes, giving null input to SBJ for
the state nodes not influenced by action nodes directly.
Vink(A,B
; ink(A,B) ‘os
path(A,B)
Figure 6 Link and path
Interconnecting the causal relations obtained’ above, the causal-loop
diagram is prepared. The process can be automated by using PROLOG or LISP,
because those programming languages have sucha function inherently. In
causal-loop diagrams, the links between the nodes related indirectly should
be eliminated, otherwise the number of links will increase, resulting in a
complicated diagram. Some of the knowledge required for eliminating such
inappropriate links are represented as the following heuristic rules (see
Figure 6).
Rule 1: If Link(A,B, _,_),
and: Path(A,B),
and Action(A,_, 4 »_)>
and Action(B,. 5 5 »_)+
and Path(A,B) does not contain Action node except A and B,
then Link(A,B,_,_) is eliminated. (5)
Rule 2: If Link(A,B,_,_),
and Path(A,B),
and State(A,_),
and Action(By_5_»»_)+
and Path(A,B) does not contain Action node except B,
then Link(A,B,_,_) is eliminated. (6)
Rule 3: If Link(A,B,
and Path(A,B),
and State(A,_),
and State(B,_),
and Path(A,B) does not contain Action node,
then Link(A,B,_,_) is eliminated. @)
Rule 4: If Link(A,B,_,_),
and Path(A,B),
and Action(A,OBJa,_»__)»
and State(B,SBJb),
and OBJa is not SBJb,
and Path(A,B) does not contain Action node except A,
then Link(A,B,_,_) is eliminated. (8)
where, Path(A, B) is a sequence of links (excluding Link(A,B)), and the each
link in the path is directed toward node B and away from node A. Those rules,
of course, are not absolute; however, the causal-loop diagram resulting from
applying these rules illustrates a clearcut structure of the system. In order
to seize the whole structure of the system and the detailed structures of the
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 311
sub-systems, it is required to divide the causal—loop diagram into parts
according to the action node (see Figure 11, for reference).
2.2. Developing Computer Model
In developing a computer model to simulate the system behaviour, it is
generally helpful to prepare a flow-diagram by refining- the causal-loop
diagram. The flow-diagram ‘provides additional insight into the system
structure. The necessary step in refining the causal-loop diagram into the
flow diagram is the identification of system levels, rates, auxiliaries, and
parameters ( in this paper, "parameter" means the system boundary node ). The
knowledge required for identifying system levels, rates, etc. is represented
as the following rules.
Rule 5: If Action(A,_» s»_)» .
then Node A is Rate. (9)
Rule 6: If Link(A,B,_,_) does not exist,
and Node B is not Rate,
then Node B is Parameter. (10)
Rule 7: If Link(A,B,_,_),
and Action(A,OBJa,_, »_),
and State(B,SBJb),
and OBJa is SBJb,
then Node B is Level. (1)
Rule 8: If Node A is not Parameter,
and State(A,_),
and Node A is not Level,
then Node A is Auxiliary. (12)
The computer model is developed by formulating the flow-diagram, In the
following, the process of generating a computer program is described:
1) Create an abbreviated name for each variable (node) provided that
one~to-one correspondence is found between them,
2) Write the equation for each variable as a function of the variables in
the upper stream. Because of its conventional/standardized. format, the
rate-level equation is automatically generated referring to "SIGN"
(positive or negative) in Eq. (4). As for parameters, each of them is
expressed as a function of time. The equations are represented
following a computer language statement, e.g. BASIC.
3) Save these equations into a file.
4) Specify the initial value of each system level. They are expressed in
the form of equalities.
5) Save these initial values into a file.
6) Merge the equation file with the initial-value file, and complete a
computer program adjusting properly:
2.3. Organizing Acquired Knowledge
In order to to facilitate future modeliing work about a system related to
the ones dealt with in the past, the knowledge acquired in the systems
modelling should be accumulated in a knowledge base. The facts to be stored
in the knowledge base are the causal relations in systems and the
mathematical equations defined on variables. The knowledge base is considered
as a library/catalogue of dynamic structures of systems (Forrester 1980,
Andersen 1980). The facts about action, state, and information nodes are
codified according to some indexes. | The data structures of the stored facts
are given as follows:
312 THE 1987 INTERNATIONAL, CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
Action(AN, OBJ, AM, DM, [SYS], [IN], [SN], [EQ]) (13)
State(SN, SBJ, [AN], [EQ]) (4)
Inform(IN, [AN]) (1s)
where [ * ] is a list, one of the symbolic expression, that means a set of-.
atoms (i.e. arbitrary characters) (Chang 1973, Clocksin 1983), and EQ is the
equation for the node defined in the phase of developing computer model, In
addition, the facts about links are stored: the data structure follows Eq.
(4). In the phase of extracting causal relations, "AN" and "SYS" are employed
as keywords to retrieve the information from the knowledge base.
3. STRUCTURE OF THE PROPOSED SYSTEM
Fundamental specifications for .the expert system of system dynamics
modelling are:
1) The proposed system is intended for the users without/with technical
knowledge about system dynamics modelling.
2) Initially, the system has the rules and knowledge only about system
dynamics modelling procedure.
3) Knowledge/facts acquired in systems modelling are accumulated to
facilitate future modelling work;
A skeleton of the expert system is shown in Figure 7. The system is
composed of:
User » User interface
Working memory Inference engine
(Outside WM) CITE)
Procedural Knowledge
Fact base (PKB)
Knowledge (FB) RB MKB
acquisition
mechanism
Knowledge base ( K B )
Figure 7 Skeleton of the expert system
a) Knowledge Base (KB) - This contains the heuristic rules and procedure
knowledge about ‘system dynamics modelling and the facts acquired in
systems modelling in the past. The rules and procedure knowledge are
stored in Procedure Knowledge Base (PKB), and the facts are stored in
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 313
Fact Base (FB).
b) Fact Base (FB) -- The facts about causal relations in systems and
mathematical equations of variables are stored. The data structures of
the facts are given by Eqs. (4), (13)-(15).
c) Procedure Knowledge Base (PKB) - The rules to operate the facts are
stored in Rule Base (RB). The procedure knowledge about system
dynamics modelling are stored in Meta-Knowledge Base (MKB). PKB is the
core program of the system encoded in PROLOG.
d) Rule Base (RB) - The rules, that would be applied to the facts, are
stored. Each rule has a precondition, and it can be applied if the
precondition is satisfied. Such rules are called "production rules".
e) Meta-Knowledge Base (MKB) - The procedure knowledge about system
dynamics modelling, including the meta-knowledge/meta-rules, i.e.
knowledge/rules ‘about how to use other knowledge/rules, are stored.
f£) Inference Engine (IE) - This draws new conclusions from given facts
applying ..the rules and procedure knowledge which are loaded from PKB
-at the system starting. In addition, the facts in FB are provided to
IE according to demand. Facts are operated and stored in Inside
Working Memory (Inside WM). In this system, PROLOG interpreter plays
the role of IE (see Figure 8).
Working Memory (WM) - This is a storage area used for the facts and
other short-term information. In this system, because of the
limitation of the memory capacity, Outside WM is equipped using an
external memory device in addition to the Inside WM of IE.
h) Knowledge Acquisition Mechanism -, This extracts the knowledge about
causal relations in systems fromthe facts in WM, © and codifys them to
be stored in FB. Sometimes thé knowledge is directly imputed by the
user. my Aa hie"
Both Inference Engine and the proposed system are based on -"production
system" (Nilsson 1980, Forsyth 1986).
8
Inference engine (PROLOG)
Knowledge
Base Rules
Meta—Knowledge | PROLOG program
Outside WM Inside WM User
Figure 8 Structure of inference engine
Using the rules and procedure knowledge described in chapter 2, the
expert system generates a system dynamics model. Figure 9 shows the flow of
the system procedure. To facilitate writing equations in the phase of
formulating causal relations, the proposed system supports some built-in
functions such as TABLE, PULSE, etc. in DYNAMO.
$14 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
co
A Extract causal relations
Interconnect the causal relations,
Fact Base and
Eliminate inappropriate links
(Causal—loop diagram)
iy) Refinement
NN Tdentify system Levels,Rates.etc.. |
and
Formulate the causal relations
(Flow-diagram, and Simulation program)
a
Simulation
(System behavior)
g
Figure 9 Flow of the system procedure
4. APPLICATION EXAMPLES
In this. chapter, the following application examples are provided to
verify the‘applicability of the proposed system,
4.1. Operation Test
An opefation test is given to examine the validity and applicability of
the proposed system. Three model builders (i.e. Mr. A: a student without
knowledge about system dynamics, Mr. B: a student with a little knowledge
. about system dynamics, but he is inexperienced in this expert system, and Mr.
C: one of the authors) get the following exercise.
Exercise. :. Ecosystem of an island
Suppose an island on where hare and fox are inhabiting. an ecosystem of
the island is as follows:
Hare feed on grass, ‘and fox prey upon hare. The number of their natural
births/deaths depends on their population. The number of hare killed by fox
depends on both the fox population and the number of hare killed per fox. The
number of hare killed per: fox depends on the hare population density. The
food scarcity affects the population deeply. The number of deaths by
starvation depends on the shortage. of their food supply. The growth potential
of grassland is proportional to the area; however, the grassland area is
decreased by hare,. and the size of the area decreased depends on the hare
population. Of course, there is an upper bound of the grassland area owing to
the limited land.
It is known that the island has a large amount of deposits underground.
The digging of the deposits decrease directly the area of the grassland, i.e.
hare's habitat. Furthermore, pollution caused by tie digging will lower the
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 315.
Natural births
deaths (fox
umber of fox
Deaths by
starvation( fox)}
Parameter of natura
lbirths/deaths (fox)
latural births
births (fox)
Lt
Death by
Taraneter 0 Istarvation(hare:
larass food
onsumpt ion
5:
Eicareter of natural
irths/deaths(hare) deaths (hare)
(b) Model B. prepared by Mr.B
aging o
leposits
Ipper bound 0 Ipper bound o ‘ {Upper bound o
rassland_areal are habitat fox habitat
"opulatior ‘opulat ion]
rowth 0 laturaT births|,_lensity density latural births
yrassland_are: (hare) hare) \(fox), (fox)
rasstand areal Number of hare] jumber_of fo3]
‘0d for] Pray of]
re fox
latural deaths} latural deaths]
jeaths by] [(hare) jeaths by (fox)
starvation Istarvation|
hare) (fox)
(c) Model C prepared by Mr.C
Figure 10 Causal-loop diagrams for an ecosystem
316 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
“estes qetaaneesgiseraraeme store esteeatene eee
* Causal Links (Sub-System) *
re
{SHURE Hiatt intial teats
feate sub-svsTew[ 1] #ff4#
digg ing_of_deposits
upper_bound_of_grassland_erea
upper_bound_of hare habitat
NODE
NODE
ce)
NODE = upper_bound_of_fox_habitat
LINK = digging_of_deposits -> upper_bound_of_grassland_area
LINK = digging.of deposits -> upper_bound_of_hare_habitet
LINK = digging_of deposits -> upper_bound_of_fox_habitat
#HREE SUB-SYSTEM[2] #448
NODE = natural births hare
TRE HEHEHE HEH 2
sana Saugal Links (Total-system) | * ani pavel
"osha em neath ei ete
NODE = SUB-SYSTEM[ 1] chare
NODE = SUB-SYSTEM[2]
NODE = SUB-SYSTEM[ 3} te parte
N = J qatura: irths: re
HODE >= 'SUB- SYSTEM A] aatural_deaths_hare
LINK = SUB-SYSTEM[2](NODE=18) ->. SUB-SYSTEM[3](NODE- 17) |Jeaths_by_starvation_hare
LINK = SUB-SYSTEM[2](NODE= 4) -> SUB-SySTeM! alwonr= 11) [>rav_of_fox
LINK = SUB-SYSTEM[4](NO
LINK = SUB-SYSTEM[2](NO | stitsnitinniniesiecnmnit REAR
LINK = SUB-SYSTEM[2](NO Paths between Sub-Systems *
LINK = SUB-SYSTEM[2](NO | itisteesassnerecrsareinea ite iritirttitasbsitestitietde
LINK = SUB-SYSTEM[3](NO
LINK = SUB-SYSTEM[1](NO | PATH = SUB-SYSTEM [1]--->SUB-SYSTEM [3]
LINK = SUB-SYSTEM[1](NO (1)upper_bound_of_grassland_area-—->(16)growth_of_grassland
LINK = SUB-SYSTEM[1](NO- ~
PATH = SUB-SYSTEM [1]---»>SUB-SYSTEM [2]
(2)upper_bound_of_hare_habitat—-->(18)population_density_hare
SHhtneeenamennieenaeanananntemennnace | SUBSYSTEM [4]
* Sub-Systems and Nodes x |iabitat--->(17)populat ion_dens ity_fox
ESHEETS ITE ESE 4
SUB-SYSTEM [3].
#4nEe SuB-sysTEN (1) ##4HF 11)food_for_hare
--~digging_of_deposits sissystenta
~upper_bound_of_grassland_area
upper bound-of hare habitat MAddeaths by starvation fox
SuB-SYSTEM [2]
-upper_bound_of fox_habitat
si SUB-SYSTEM [2] ##### 10)deaths_by_starvation_hare
8---natural_births hare .
Timbar_Bf hare” SUB-SYSTEM {2}
~natural_deaths_hare 5)pray_of_fox H
~deaths_by_starvation_h
es a aaa SUB-SYSTEM [3]
-population_density hare fares) food_for_hare
#EHHE SUB-SYSTEM (3) ##RRE
1i---food_for_hare
5---grassland_area
16---growth_of_grass land
Figure 11 Example of the system's output (model C)
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 317
growth potential of the grassland.
Develop an ecosystem model of the island to examine the effect of the
digging of deposits.
Modelling of the ecosystem structure is carried out using the proposed
system. Figure 10 shows the results illustrating the system's output with
causal-loop diagrams. Though model A and model B are fairly analogous, there
are many difference in the three models with regard to the level of the
problem recognition. However, the each model succeeded in extracting the same
critical concepts and feedback loops in the system; so that the proposed
system is not altogether worthless. If the model builders enter on: the phase
of system formulation, then those models will be refined fairly well. It
should be praised that Mr. A detected the causal relation about fox and
hare's excretions. Figure 11 shows an example of the system's output for
model C.
4.2, Modelling of Maritime Industries
On account of complicated international and economical circumstances,
shipbuilding and shipping industries in the developed countries are going
through their serious recession (Nersesian 1981, Taguchi.1986, Kishi 1986).
In order to find suitable steps, it may be appropriate to’ employ system
dynamics models in the policy analysis. In this section, a system dynamics
model for maritime industries is presented using the proposed system.
Gi} taransport
Average transport |_-»{demand
Distance Freight
7 bates
(Order tonnagel
ito construct.
Demand & supp]
balance
Stock tonnage
rata to construct
‘otal tonnage} Construction]
lof vessels (3 tonnage ~~
Figure 13 Causal-loop diagram for tanker fleets system
(SD Research Society in JMRI 1977)
Extraction of the causal relations of the system are pursued by the
authors, Maritime indus es are composed of shipping, shipbuilding, and port
& harbour. And further, for example, shipping is divided into specialized
markets (tanker, bulker, container, etc.) (Nersesian 1981). Figure 12 shows
a causal-loop diagram for the tanker fleets system obtained using the
proposed system, and the identified system levels, rates, etc. Japan
Maritime Research Institute has already provided a similar model as shown in
318 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
4forder tonnagebe10 [Tonnage into
to construct | ]@ shortage after
release of
Taying-up
1Btock" tonnage AT
[to construct 4
Tom
|a_ shortage
5{Construction i
[tonnage 16 [Average ] inftransport]__ &
speed of Working days
vessels | Loapscity
Scrapping of 2 Operating]
operating vessels| tonnage 16.
Marine transport
W demand
Excess tonnage]
after laying-
lup scrapping
15
1efiverage age ‘afRetease of] —
lof vessels faying-up_} Keying-upj*—[Excess_tomage
7 {Scrapping of 3fLaying-up] 13[Excess tonnage
laying-up vessels }—\cm4 tonnage after laying-up
sme LEVELS Se a
scrapping_of_laying_up_vessels
Node No, == 1 ship
Node name Name -- stock_tonnage_to_construct
subject ---- ship 8
release_of_laying_up
Node No. == 2 Object ship
Node Name -~ operating_tonnage
Subject ---- ship Node No. --- 9
Node Name -- laying_up
—3 Object ----- ship
Taying_up_tonnage
ship so RUKILIARIES #4
Node No. ~~~ 10
Node Name -- tonnage_into_#
Heeme RATES Het shortage_after_release_of_laying_up
Node No, ~-- 4 Node No. --- 11
order_tonnage_to_construct Node Name ~~ excess_tonnage_after_laying_up_scrapping
ship
5
construction_tonnage
ship sek PARAMETERS, sth
6 Node No. --- 18
Scrapping_of_operating vessels Node Name —~ working_days
ship
Node No. =-- 19
Node Name ~~ marine_transport_demand
Figure 12 Causal-loop diagram for tanker fleets system
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 319
Figure 13 (SD Research Society in JMRI 1977). Although the two models differ
in their elaborateness, the central structures of the causal relations in the
models are equivalent. Behaviour of the each model is simulated, and the
results are shown in Figure 14 for reference.
t
——Authors' model Total tonnage
-JMRI'S pode? Laying-up tonnage
Order tonnage to construct , Laying-up tonnage (x10°OW)
,mStatistics Q Order tonnage to construct
F
Total tonnage (xt0%Du)
Forecastin«
n % by a
“SM 75 30 cc by z 2000
Figure 14 System behaviour
5. CONCLUSIONS
This paper is concerned with an expert system to support system dynamics
modelling. The application examples are provided to verify the applicability
of the proposed system. The results are summarized as follows:
1) The process for system dynamics modelling is investigated. Some rules and
procedure knowledge for preparing system dynamics models are presented, and
the data structures of causal relations are given in the form of predicate
logic formula.
2) An expert system for system dynamics modelling is designed including
the knowledge base about causal relations.
3) The results of the application examples are in the following: i) The
proposed system effectively supports the system dynamics modelling not only
by the user with little technical knowledge about system dynamics but also by
experienced user. ii) In eliminating inappropriate links in the causal-loop
diagrams for large systems, the proposed system needs some heuristic rules to
avoid the problem of infinite number of path combinations,
320 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
REFERENCES:
Andersen’ “D.F., and Richardson, 6.P. (1980), "Toward a Pedagogy of System
Dynamics", TIMS Studies in the Management Science, No. 14, North-
Holland, pp 91-106
Axelrod R.(ed.) (1976), Structure of Decision: The Cognitive Maps on
Political Elites, Princeton University Press.
Chang, C.L., Lee, R.C. (1973), Symbolic Logic and Mechanical Theorem Proving,
Academic Press.
Clocksin, W.F., and Mellish, C.S. (1983), Programming in Prolog, Springer-
Verlag.
Elzas, M.S., Oren, T.I., and Zeigler, B.P. (eds.) (1986), Modelling and
Simulation Methodology in the Artificial Intelligence Era, North-
Holland.
Forrester, J.W. (1961), Industrial Dynamics, MIT Press.
Forrester, J.W. (1980), “System Dynamics - Future Opportunuties", TIMS
Studies in the Management Science, No. 14, North-Holland, pp 7-21
Forsyth, R., and RADA, R. (1986), Machine Learning: Applications in Expert
Systems and Information Retrieval, John Wiley & Sons.
Kishi, M., Nakayasu, H., Noda, K., and Taguchi, K. (1986), "Cognitive Map and
System Analysis of Maritime Industry", in: Hubner, H. (ed.), The Art and
Science of Innovation Management, Elsevier Science, pp 385-399
Nersesian, R.L. (1981), Ships and Shipping: A Comprehensive Guide, Penn Well.
Nilsson, N.J. (1980), Principles of Artificial Intelligence, Springer-Verlag.
Nolan, \P.J., and McCarthy,.M.A. (1986), "AI Frame-Based Simulation in System
Dynamics", in: Sriram 'D., and Adey R. (eds.), Applications of Artificial
Intelligence in Engineering Problems, Vol. 1, Springer-Verlag, pp 527-
538
Pugh, A.L. (1976), DYNAMO User's Manual, Fifth Edition, MIT Press.
Roberts, N., Andersen, D.F., Deal, R.M., Garet, M.S., and Shaffer, W.A.
(1983), Introduction to Computer Simulation; A System Dynamics Modeling
Approach, Addison-Wesley.
SD Research Society in JMRI (1977), "SD Model for Shipping and Shipbuilding -
Tanker Market" (in Japanese), Journal of the Japan Maritime Research
Institute, No. 130, pp 5-21
Starr, P.J. (1980), "Modeling Issues and Decisions in System Dynamics", TIMS
Studies in the Management Science, No. 14, North-Holland, pp 45-59
Taguchi, K., Kishi, M., Nishimura, M., Nakayasu, H., and Hasegawa, N.(1986),
" "Normative Forecasting Model for Promoting National Shipping", Proc. of
3rd Latin-Ibero American Congress on Operations Research and Systems
Engineering.
