Simulating the man-in-the-loop
- PDF / 2,386,046 Bytes
- 5 Pages / 589.56 x 841.92 pts Page_size
- 38 Downloads / 175 Views
(ES), which model the decision-maker. in this article, Terry Williams proposes a combined route, using a VIS to elicit behaviour, then ES to emulate the behaviour, then embedding the ES within the simulation. The ,nethodolois illustrated by a case-st udy of replenishment at sea in the Royal Navy. -ooüoo-
Copyright
/996 (}percuional Research Society.
Simulation, VIS and expert systems Simulation is a well-known and widely-used technique. For years it has come near the top of regular lists of 'most used techniques'. In the early days of simulation, the technique proved a powerful means
of analysing mechanistic systems. However, increasingly systems being modelled include one or more human decision-makers, and their actions determine the course of events being modelled production systems have controllers, armies have commanders, stock-systems have expediters Without modelling this intelligent behaviour, a
17
OR insight Vol 9 Issue 4 October - December 1996
simulation can be naive and possibly even useless -
tional understanding to the decision-maker for that particular realisation of the probabilities.
the client as too simplistic. This article is concerned with situations where there is a single sophisticated decision-maker, who has to be modelled in detail for the simulation to be credible and useful.
Let us turn the to the second approach suggested above, (ii), which is to model the decision-maker
it will be incapable of representing a realistic sequence of events and reactions, and will be seen by
within the simulation. This would often, for modelling a sufficiently sophisticated decision-maker, lead to an Expert System (ES) (if the demands of requisite
modelling required that level of modelling). The
O'Keefe and Roach (1987) discuss this point, saying that simple approximations (cg simple probabilities)
combination of Artificial Intelligence (Al) (specifical-
ly in this case Expert Systems) has been growing
have often been used instead of modelling behaviour, thus falling into the over-simplification trap.
rapidly for some years: Elzas (1986) feels it obvious that simulation and Al should be coming together; O'Keefe (1986) gives a taxonomy of expert systems and simulation combinations, which includes a configuration in which an expert system is embed-
They identify a dichotomy; either:
the simulation must have access to the decision-maker to make decisions where necessary; or
ded within the simulation. This is perhaps more appropriate for more closed systems, where there are generally clear boundaries, clear choices, and
the decision-maker must be modelled.
clear objectives.
How actually to carry out such an embedding, of an ES into a simulation, is more contentious. Russell et al (1992) say that "unfortunately, there seems to be almost as many solutions to the problem of integrating simulation and artificial intelligence as there are
Looking at the first of these lines, (i), this is the line
followed by Visual Interactive Simulation (VIS).
Visual Interactive Modelling (VIM), and parti
Data Loading...
