Environmental Engineering Reference
In-Depth Information
With this set of forces and laws for solving problems, a decent number of
bars should be able to be split up into PACO agents and added to the system.
Hereafter, they will align themselves in a roughly optimal way or stay in motion,
trying to seek their goals.
Apparently, the real challenges of applying the PACO paradigm to an
agent-based control system like the one in this case study is designing and
fitting the forces used for interactions.
The agent group, which spawns from the creation of agents for a single bar
of items, is not modeled or implemented as a sole entity in the system. Thus, no
overall goal or intensions of the group can be directly implemented, but must be
realized through the aggregation of subgoals met by the agents within the group.
The tension appearing inside a group due to the spring forces of the agents
can, to some degree, lead to competitions among agents within a group, but the
social laws make it easier for the system to reach an equilibrium and dampen
the interagent tensions. Particularly, laws 1 and 3 are added to cope with these
side effects of the basic forces. Law 1 simplifies the process of attraction and
stabilizes the movements of a successor agent to its predecessor, due to the
expansion of the current time slot for an agent in a bath, if it is too hard to pack
the schedule for a bar tighter. Note that the plan for each bar at the end must
form a consecutive sequence of visits to baths as the cranes move bars from bath
to bath, because the system has no spare slots that temporarily can hold a bar.
Law 3 more directly compacts the plan of a group and increases robustness in
the coordination process.
Law 2 is important as well, even though it is orthogonal to the agent groups.
It adds flexibility by minimizing the slot time requested by an agent. It is not a
direct coordination mechanism between agents from different groups, but allows
some mutual impact on their actions.
The most challenging part of optimizing the overall plan for the system is to
decide when and how conflicts between agents should be solved. No method or
measurements exists to validate if a current configuration is optimal or jumps
between the agents should be handled. Laws 4 and 5 direct the trivial conflicts to
be handled without contracting classic local optimization principles. Law 6 serves
to dampen intergroup tensions, especially to avoid oscillating shifts between
agents from different groups with interest in the same bath.
13
RESULTS
The agent-based approach for the control system can be tested to see if it can
create valid plans for the system, which can be done by measuring a satisfactory
rate for an agent group. A fully valid plan would have a 100 percent satisfactory
rate, which means that for a given bar, all visits to baths in the recipes comply
with the minimum and maximum time frames and that moves between two
Search WWH ::
Custom Search