Environmental Engineering Reference
In-Depth Information
combine partial results to find the overall solution to the problem. The different
knowledge sources each represent a different aspect or hypothesis on the prob-
lem. These are connected to the blackboard and can modify and update the
current solution through the shared memory, which contains the definition of the
problem.
In the following years, research led to new approaches for distributed problem
solving. The contract-net protocol introduced by Smith (1979) was a turning
point in DAI. In contrast to the blackboard model, the contract-net protocol
has a managing node, which through messages propose a task to the different
knowledge sources, which each bid on the task. The manager decides which
(could be several) of the contracting nodes can carry out the task and eventually
return the result to the manager.
Hewitt was interested in the modeling aspects of distributed problem solv-
ing and introduced the actor model (Hewitt 1977). Actors are computational
entities with both a script that defines the actions and a list of other actors it
can contact — the so-called acquaintances. In the model, actors are awakened
when they receive messages from others actors. The actor then runs its script,
will die, and is subsequently removed by the garbage collector. During exe-
cution of the script, it can both spawn new actors and send messages to its
acquaintances.
Given the concepts of message passing and well-defined behavior through
the action script, the actor model was a natural predecessor to the multiagent
paradigm.
Multiagent systems are appropriate for studying and managing dynamical and
heterogeneous systems. They are an approach to handling the increasing com-
plexity of centralized systems by breaking them into simpler tasks, which also
give a more natural modeling approach.
Starting with the simplest nonintelligent agents, there exist no commonly
agreed definition of an agent, but it is generally accepted that it involves some
kind of autonomy, which means that the agent is allowed to choose its own
action. Also, the notion of being in an environment is central to agents, as they
base their actions on sensory impressions from the environment, which could
be either physical or virtual environments. Thus, some sort of input function or
perception unit is required for an agent. One of the most cited definitions of an
agent is given by Wooldridge:
An agent is a computer that is situated in some environment, and that is capable
of autonomous action in this environment in order to meet its design objectives
(Wooldridge 2002).
An agent can be as simple as your heating thermostats, but intelligent agents
are the most interesting to be studied in both agent and multiagent systems.
Figure 3.3 presents the classic illustration of an agent situated in its environment.
Search WWH ::
Custom Search