System Dynamics '95 — Volume II
TRANSPORT FORECASTING BASED ON ARTIFICIAL LIFE CONCEPT
KISHI Mitsuo, HOSODA Ryusuke, YAMADA Tomoki, FUNAHASHI Hiroki, and
MATSUMOTO Ippei
Department of Marine System Engineering, College of Engineering,
University of Osaka Prefecture, Sakai, Osaka 593, JAPAN
Abstract
A system dynamics model based on Artificial Life (AL) concept is proposed for transport forecasting. The
proposed model focuses on the economic behaviour which emerges out of the interactions among
individual local objects, i.e. economic units. That model is merely a large aggregation of simple programs
which specify how the local objects react in the environment. Application examples are provided to
illustrate the applicability of the proposed model.
1. Introduction
The forecasting should begin preparing a variety of pictures, herein called "rich pictures"
(Checkland, 1981), for the future environment. The next step is analyzing the system behaviour
of interest for individual rich picture. Basically, qualitative analysis is more preferable than
quantitative analysis in long-term forecasts. The extrapolative forecasting, which depends on
actual record of the past, does not cope with drastic changes in environmental conditions. In
order to study the "future-as-it-can-be," we desire a system dynamics model which can adapt to
those rich pictures.
Artificial life (AL) is the study of man-made systems that exhibit behaviour characteristics
of natural living systems (Langton, 1988). The Artificial Life focuses on the system behaviour
which emerges out of the interactions among micro/local objects. Thus, the AL is concerned with
system dynamics. The system dynamics model based on the AL concept can flexibly adapt itself
to the given environment, since that model is merely a large aggregation of simple programs
which specify how the local objects react in the environment.
In this paper we propose a forecasting model based on the AL concept for transport network
growth. This model is an artificial world where the local objects/units, i.e. people, resources
developers, energy factories, products factories, transporters, traders, banks, and central/local
governments, act for individual purpose. The AL-based model will present us a variety of
scenarios in transport planning. Application example of East-Asia transport network growth is
provided to illustrate the applicability of the proposed model.
611
Parallel Program
2. System Dynamics
2.1. System dynamics model
The objective of system dynamics modelling is to obtain the mathematical description, i.e. state
equations. A discrete time model can be written
X(t+1) =£(XO), 1), qd)
where X(t) denotes the state vector at time t, I(t) denotes the input vector, and f is a function
vector. A continuous time model can be written
dX/dt = (X(t), 1D). (2)
In general, the outputs from a system are related to the both variables of state and input, so that
we may write
¥() =g(X(), 1), QB)
where Y(t) is the output vector, and g is a function vector. In this paper the both variables of X
and I are called local variables or local objects.
When analyzing a system related economical, social, technological, and ecological issues, we
usually employ a discrete time model, e.g. econometrics model, etc. System-Dynamics proposed
by J.W. Forrester is one of the discrete time models (Forrester, 1961).
The process of the System-Dynamics modelling is: i) Extracting the causal relations in a
system, ii) Preparing a causal-loop diagram, iii) Transforming the causal-loop diagram into a
flow-diagram by identifying system levels, rates, auxiliaries, and parameters.
2.2. Classical system dynamics model
The interactions among local variables determine the model behaviour. In the System-Dynamics
or in the econometrics, usually it is not so difficult to manipulate the model behavior of interest.
Such manipulations in the model behaviour are possible under the following conditions:
a) The hierarchical differences among the output variables and the local variables are small.
b) The both functions of f and g in Eqs.(1)-(3) are explicit, especially linear.
c) The function f is not chaotic.
On these conditions we have a well-structured model whose behaviour may sometimes be
explicitly operatable. This is the "classical model."
The interestingness of system dynamics analysis is to find unexpected behaviour which
emerges out of the interactions among simple objects. For the "emergent behaviour," the system
dynamics model should not be constrained by the conditions a)-c).
3. System Dynamics Based on Artificial Life Concept
3.1. Artificial Life
Natural life emerges out of the interactions of a great number of cells. Artificial Life (AL)
employs the following synthetic approach to the study of life-as-it-could-be (Langton, 1988): i)
Creating simple rule-governed objects, ii) Constructing a large aggregation of the objects, iii)
Generating life-like behaviour as the result of the local interactions among individual objects.
The Artificial Life focuses on the problem of behaviour generating. Thus, the AL is concerned
612
System Dynamics ‘95 — Volume II
with system dynamics. For AL systems, the ongoing dynamics is the behaviour of interest rather
than any final state. System dynamics models based on the AL concept have the following
features:
A) They are large aggregations of simple objects or programs.
B) Each program specifies how an object reacts in its environment, including interactions
with other objects.
C) There are no programs which directly dictate the behaviour at levels higher than the
individual program.
The AL-based system dynamics model will not satisfy the conditions of a) or b) in 2.2.
Object-oriented programming is useful for AL systems. This computer-based modelling
methodology positively requires "analysis by simulation."
Cellular automata, which consist of cells arranged in a regular lattice, are models for AL
(Tamayo, 1988). The state of each cell takes discrete values, and a discrete time model is
introduced to the state transitions. The new state is defined as a function of its own previous
state and the state of the neighbourhood. Figure 1 illustrates a sequence in the evolution of a
simple two-dimensional model. We can see emergent patterns and the diffusion.
(a) t=0 (b) t=18
(©) t=55 (d) t= 293
Figure 1 Evolution of a two-dimensional cellular automaton
613
Parallel Program
3.2. Ricardian model based on AL concept
Classical school economists are the people of "analysis by thinking." We are surprised at their
clear thinking. David Ricardo explained the doctrine of comparative advantage, which is the basis
of the case for specialization and for freedom of trade, by means of the following example
(Hartwell, 1971).
Suppose that Portugal has an absolute advantage in the production of both commodities of
wine and cloth as shown in Table 1. If the post-trade exchange rate between the commodities is
reasonable, then it would be advantageous for Portugal to employ her capital in the production
of wine, and to export wine in exchange for cloth. Thus, England would import wine by the
export of cloth despite her absolute inferiority. The both countries can simultaneously gain from
trade.
Table 1 Productivity of commodities
Labour hours required to
produce
1 gallon 1 yard
wine cloth
Portugal 80 90
England 120 100
Modern economics is not reliant on the Ricardo’s theory any longer; however, his theoretical
analysis does not lose its value. Even if it is difficult for us to tower above Ricardo in the
theoretical analysis, we can rank beside him by taking a different approach, the analysis by
simulation.
Simulation analysis can reveal the characteristics of an economical system. For example, the
continuous time model, i.e. differential equations model, used to be introduced to illustrate the
Ricardo’s theory (Caravale, 1980). Now we present a Ricardian model based on the AL concept.
The AL-based model is an artificial world which consists of England and Portugal. Wine
producers, cloth producers, traders, and consumers in both countries act pursuing individual
profit. The producer wants to increase the production considering the demand for the products.
The trader wants to increase the volume of business considering the supply and demand for the
products. The trader has dealings with national producers and foreign traders. The consumer
demands constant volume of wine and cloth for a given period of time. The consumer wants to
buy cheaper products, i.e. national or imported goods.
Given the post-trade exchange rates between the commodities and the initial conditions, the
behaviour of the Ricardian model can be simulated. Figure 2 shows a result. We recognize the
specialization there.
614
System Dynamics '95 — Volume II
Cloth
Production
Wine
OF, :
L 1
Time
Figure 2 Products specialization in England
4, Transport Network Growth Model Based on AL Concept
4.1. Forecasting model
Since the transport depends on national/international economy, a demand forecasting model for
transport should be coupled with an economic growth model.
Extrapolative forecasting models, which explicitly dictate global behaviour without
considering the interactions among local objects, can not cope with drastic changes in
environmental situations. Forecasting models considering the system structure possibly can adapt
to the drastic change. Especially, the AL-based model can adapt itself to the given environment,
since that model is merely a large aggregation of simple programs which specify how the local
objects react in the environment.
AL-based system dynamics models, which generate global behaviour implicitly but not
explicitly, will present us a variety of scenarios in transport planning.
4.2. Economic growth model
Production, distribution, consumption, and investment are the main activities in circular flow of
economic system. Finance and government administration concern those activities.
The prototype model for economic growth based on the AL concept is an artificial world
which consists of N countries. Figure 3 illustrates the model framework. In every country the
local objects/units, ie. people, resources developers, energy factories, products factories,
transporters, traders, banks, and central/local governments, act for individual purpose as follows:
a) People (Pch) - People unit Pch (the subscripts c and h denote country and identification,
respectively: c=1,...,N, h=1,....NPc) is employed by an enterprise unit. It consumes energy
and products for people. It can change its employment and/or residence for a large salary.
Its expense is in proportion to its income. It wants to buy cheaper products. Its
615
Parallel Program
propensities to save and to change employment/residence are influenced by the people
units’ propensities in its immediate neighbourhood.
b) Resources developer (RDcri) - Resources developer unit RDcri (the subscript r denotes
kind of resources, r=1: for energy, r=2: for products, and i denotes identification:
i=1,..,NRDc) outputs resources by inputting labour, energy, and capital equipment. It
supplies resources to factories. It wants to increase the profit. The substitution between
capital and labour is possible.
c) Energy factory (EFcj) - Energy factory unit EFcj (the subscript j denotes identification:
j=i,....NEFc) outputs energy by inputting labour, energy, energy resources, and capital
equipment. It supplies energy to people and enterprises. It wants to eliminate the deficit.
The substitution between capital and labour is limitational.
d) Products factory (PFcpk) - Products factory unit PFcpk (the subscript p denotes kind of
products, p=1: for people, p=2: for enterprises, and k denotes identification: k=1,...,NPFc)
outputs products by inputting labour, energy, products resources, and capital equipment.
It supplies products to people and enterprises. It wants to increase the profit. Its
willingness to invest is influenced by the products factory units’ willingness in its
immediate neighbourhood. The capital equipment is composed of the products for
enterprises. The substitution between capital and labour is possible.
e) Transporter (TPcl) - Transporter unit TPcl (the subscript 1 denotes identification:
1=1,....NTPc) conveys resources and products by inputting labour, energy, and capital
equipment. It wants to eliminate the deficit. The substitution between capital and labour
is limitational. See 4.3.
f) Trader (TDcm) - Trader unit TDem (the subscript m denotes identification: m=1,...,NTDc)
plays an intermediary role in the international/domestic trade of resources and products.
It wants to increase the profit.
g) Bank (Bcn) - Bank unit Ben (the subscript n denotes identification: n=1,...,NBc) takes care
of money, i.e. receiving and lending. People and enterprises deposit money in it, and
enterprises raise funds from it. It wants to adjust saving to investment by interest rates
operation.
h) Local government (LGcq) - Local government unit LGcq (the subscript q denotes
identification: q=1,....NLGc) engages in local public services, ie. giving subsidies to
energy factories, and aiding poor people. It wants to eliminate the deficit financing.
People and enterprises shall be liable to local taxation.
i) Central government (CGc) - Central government unit CGc engages in national public
services, i.e. giving subsidies to transport liners, and aiding poor local governments. It
improves ports and transport routes by inputting labour, energy, and products for
enterprises. It wants to eliminate the both deficits of trade balance and financing. It
imposes customs on imported products and resources. People and enterprises shall be
liable to national taxation. The central government sometimes plays a centralized
controller against AL concept, such as revaluing/devaluing foreign exchange rates,
opening/closing ports/routes/economy, etc. The international exchange market is not
considered at present.
In order to keep and increase the productivity, enterprise units invest in their capital equipment.
Moreover, they have to invest against environmental pollution. Heavy pollution inactivates people
units; so that the labour productivity will decrease. And, the local government unit will order the
enterprise units to invest more against pollution or to stop operation.
616
LI9
Pitt
Foreign
Countries
: Products/Energy
: Resources
: Labour
: Money
nt) oe
H Local Central }
t government government {
Gea CGe Administration 1
{ t 4 '
rt ut
i Ht
i bil
1 Resources itt
13 I developer | | it
tt 1 rit
| 114
Vy i itd
1 bcxxczzp| EF oj
Pech — Production } |
i Object-economy | |
| Investment A) Saving
| Bank |
1 i
L Ben Market economy }
Cin CCC CCC
Figure 3 Framework of economic growth model
TI SwINjOA — C6, SoreUXg Wa\skg
Parallel Program
People units increase in population on a long time scale. Active enterprise units increase also
in population, but on a short time scale. Each unit performs according to its own strategy. It is
possible for the unit to change its strategy at every iteration stage. The poor unit follows the
strategy of a rich unit in the neighbourhood; however, such a mimic habit does not necessarily
guarantee success. Mutation is occasional random alteration of the strategy (Goldberg, 1989).
Mutation rates are appropriately small. The mimicry and mutation are the built-in mechanism of
the natural selection in this model.
4.3. Transport network growth model
Transport network should grow according to transport demands. A transport route consists of
links. A link connects an origin node and a destination node, where the node means a port or a
freight station in a production complex. The central government unit and transporter units want
to invest in the link congested with cargoes. Barriers to entry into the congested link are open
to outside transporters.
Trader units want to transport the commodity as economically as possible. Thus, the freight
competition shall be inevitable. If a trader unit can not avail the direct freight between two nodes,
the trader will search for an indirect and economical freight, if possible.
Time preference for commodities is not considered at present. However, if the commodities
are classified more‘sophisticatedly, then trader units have to take into account the time preference
of individual trade commodity. And transporter units have to provide rapid transport services. The
mutation and mimicry will also play the important role in the distribution revolution.
5. Application to East-Asia
5.1. East-Asia model
The transport network growth in East-Asia is investigated. The market economies along the
Asia-Pacific Rim have registered extremely rapid growth, and the pressure for improving the
transport network is building up especially in China.
One of the sources of the economic growth is human resources development (Ogawa, 1993),
e.g. population growth, higher standard of education, higher morale, etc. The population growth
tates have been declining very rapidly in the past quarter century in East-Asia. Hence, we assume
no demographic change in our model. Educational development is not considered here.
The artificial world is composed of China (the east side), South Korea, North Korea, and the
outside. The countries trade with each other and also with the outside through the boundary;
however, North Korea is closed to the other countries except for China.
Suppose that the countries can import resources and products from the outside as they wish,
and that the international price of the outside commodity is fixed. On the other hand, exports
to the outside are restricted within the limits.
5.2. Numerical example
Setting the initial conditions in consideration of the present state, the behaviour of the artificial
world is simulated. A people unit corresponds to a population of million.
One of the current topics of East-Asia is the North Korea problem. It is highly probable that
North Korea will open the country in the not very distant future. Although we have not yet
realized an endogenous mechanism of the political revolution in our model, we can investigate
618
System Dynamics '95 — Volume II
the impact of the revolution by using the model, for example the impact upon the transport
network. The trade amount between China and South Korea is rapidly growing, so that the
transport modal shift from the Yellow Sea (Hwang Hai) route to the transit route via North Korea
might occur after the revolution.
The AL-based model produces various simulation results. Figure 4 illustrates an interesting
tesult. We recognize the transport development between the northeast China and South Korea via
North Korea, as expected. Nevertheless, the Yellow Sea route does not decline. This means that
the transport network growth recreates new demand for transport. At least it may be said that the
AL-based model is more imaginative than the authors of this paper.
r T T T
SK-neC
(Yellow Sea route)
North Korea revolution
SK-neC
(North Korea route)
Transport between countries
L Ll in
? Year
Figure 4 Impact of North Korea revolution
6. Conclusion
NK : North Korea
SK : South Korea
neC_ : the northeast China
This paper is concerned with the transport forecasting model based on Artificial Life concept.
The results are summarized as follows:
1) The system dynamics methodology based on the AL concept, which positively requires
“analysis by simulation," is proposed.
2) The Ricardian model based on the AL concept is proposed, and the simulation example
of the comparative advantage is provided.
3) The prototype model for transport network growth based on the AL concept is proposed,
together with the economic growth model.
4) Application example of the transport network growth in East-Asia is provided to illustrate
the applicability of the proposed model.
A part of this work is financially supported by a Grant-in-Aid for the Scientific Research, the
Ministry of Education, Science and Culture of Japan. All the computations were processed by
using the Sun SPARC system.
619
Parallel Program
References
Caravale, G.A., and Tosato, D.A., (1980)
Ricardo and the theory of value distribution and growth, Routledge & Kegan Paul.
Checkland, P.B., (1981)
Systems Thinking, Systems Practice, John-Wiley.
Forrester, J.W., (1961)
Industrial Dynamics, MIT Press.
Goldberg, D.E., (1989)
Genetic Algorithms in Search, Optimization & Machine Learning, Addison- Wesley.
Hartwell, R.M.(ed.), (1971)
Ricardo, Principles of Political Economy and Taxation, Pelican Books.
Langton, C.G.(ed.), (1988)
Artificial Life, Addison-Wesley.
Ogawa, N., et al., (1993)
Human Resources in Development along the Asia-Pacific Rim, Oxford Univ. Press.
Tamayo, P., and Hartman, H.,(1988)
"Cellular Automata, Reaction-Diffusion Systems and the Origin of Life," in: Langton, C.G.(ed.),
Artificial Life, Addison-Wesley, pp.105-124.
620
