Information Technology Reference
In-Depth Information
On the other hand, if there are agents that only use one kind of resource, the
process of its replenishing becomes impossible. When only one kind of resource is
used, it is exhausted after some time.
As a result of this situation, the agents that need that resource are not able to use
it and consequently the operation of the whole system is blocked.
Unbalanced load of the resource may be treated as the example of atypical behav-
iour (action) of the system, which should be recognized.
Therefore, it is necessary to equip the system with the mechanism for recognition
of agents that load the resources in atypical way (i.e., unbalanced).
To realize this mechanism, the agents in their model remember the order and the
type of the resource they used. The memory covers a given
n
of the previous cases of
loading the resource. Memorized cases are accessible to a given agent and all other
agents in the system.The method for recognizing the agent's atypical behaviour
consists of three stages:
The first stage is the formation of a model of atypical behaviour by each agent.
The approach based on the elements of operation of immune artificial systems
has been used here. Each agent generates randomly a model of certain length
(e.g.,
m
model), being the code of using the resources. Then it compares the
model with information stored in its memory, specifying the order in which it
used the resources. If the comparison is negative—a given model is used in further
recognition of atypical behaviours. If the result of comparison is positive (i.e., a
given agent loaded resources in the same order) then the model is rejected.
At the second stage a given agent (assessing) assesses other agents. For each
assessed agent it compares the model with the agent's behaviour it has built itself.
The result of comparison is remembered by the assessing agent. This process is
repeated by the assessing agent for every single agent of the system. As a result,
a given agent has the assessment of behaviour of all other agents in the system.
This procedure is realized for every agent of the system, therefore, each agent has
the information assessing the behavior of all other agents.
At the third stage agents send their feedback about other agents to the common
chart of the system. It is a kind of voting where every agent comments on (as a
citizen casts a vote) each one of the agents. On the basis of feedback, the resul-
tant assessment of each agent may be specified, and the agents with the worst
resultant assessment may be chosen, i.e., those whose behaviour cause atypical or
undesirable operation of the system.
The scenario presented above is cyclical except for the first stage, i.e., the genera-
tion of models which is performed only once. It allows for the control of the system
operation and receiving warnings about the appearance of atypical behaviours and
what is more, their probable sources.
The above approach to the recognition of atypical situations may be realized as
the identification of a certain behavior of the agents. It allows them to analyze the
behaviour of the system, with the use of different points of view.
In order to illustrate the above considerations, we will present the research results
carried out with the use of the simulation method on the exemplary system for
