Environmental Engineering Reference
In-Depth Information
1
0.8
0.6
0.4
0.2
1 agent group (bar)
5 agent groups (bars)
15 agent groups (bars)
25 agent groups (bars)
0
0
500
1000
1500
2000
2500
3000
Time Steps
Figure 3.28
Satisfaction rate for test with 1, 5 15, and 25 agent groups
consecutive visits are connected with no glue time (the extra time added to a
bath visit for the plan to consist of consecutive visits between the different baths).
Naturally, the computation time is also interesting as a measure for the dynamic
reactiveness of the approach, but as the results show in Figure 3.28, the agents
relative quickly move to a rather stable level of their satisfaction rate.
Valid plans are not met in all scenarios, so in order to improve the results, a
number of experiments have been conducted, which clearly improve the number
of valid plans being generated. Some of those strategies will be explained in this
section. On a meta-level, they show one of the real strength for developers to work
with agent systems: It is very easy to add or change a behavior that is local in
one agent and test whether the performance has improved, without restructuring a
central control. These strategies also give the agent a more deliberative behavior,
so by introducing such changes, the system would be more a hybrid system than
a system of purely reactive agents.
13.1
Active, Sleeping, and Locked Agents
During tests, it has generally been observed that the system could end in a
nonconverging situation, where one or more agents oscillate. Thus, a promising
approach is to give the agents a state that determines their ability to perceive the
Search WWH ::
Custom Search