Information Technology Reference
In-Depth Information
The total number of messages exchanged within each group in one step in a multi-
agent system made of two groups of 30 agents (in total 60 agents) is seen in Figure
3, left, while the target-value only information exchange is shown in Figure 3, right.
The highest number of exchanged messages is achieved by the model with target only
Fig. 3. Left: Medium total number of messages exchanged among agents for the algorithm with
all vs. partial information exchange; right: Medium total number of messages of the algorithm
with target only information exchange for 60 agents
information exchange, Figure 3. This is due to the fact that unpredicted assignment
events caused by moving targets put increased uncertainty into the system so that the
past information exchanged in each group can be misleading and might degrade the as-
signment solution. The quantity of each assignment information exchanged among the
agents in each step is at least n times less than the available information and, therefore
in each step an agent has to contact at least n times other agents repeating the whole
bidding process in sequence for each auctioned target. This large amount of information
that must be passed over the agents' communication network is a potential problem if
the latter is subject to imperfections such as delays, noise, and changing topology.
The models with all and partial information exchange show very close crossed dis-
tance results, while in the terms of the medium number of exchanged messages, the
model with target-only information exchange is more demanding than the all informa-
tion exchange model.
The most surprising result of the simulation with two groups of mobile agents, com-
pared to the results of the simulation with only one mobile agent group while the other
is static [12], is that in the former, the greedy approach gives a better global solution
than any other approach as the communication range decreases under the CTR. There
emerges a conclusion that when the environment is influenced by unknown factors and
its behavior is therefore unpredictable, it is more convenient to follow the greedy policy
where agents don't exchange the assignment information with other agents than to rely
on the information which in the dynamic environment with high probability is obsolete
Search WWH ::




Custom Search