School of Computer Science & Informatics

streakconvertingΛογισμικό & κατασκευή λογ/κού

13 Δεκ 2013 (πριν από 3 χρόνια και 8 μήνες)

69 εμφανίσεις

Richard Tynan, G.M.P. O’Hare,

Michael O’Grady & Conor Muldoon


School of Computer Science & Informatics

University College Dublin

Ireland


WSN Issues


Intelligent Power Management


Agents for WSNs


Current Approach


MAS Approach


Agent Factory Micro Edition


Resource Bounded Reasoning


Experiments


Future Work


Conclusions







Connectivity


Latency


Density


Accuracy


Energy Consumption


Is a WSN useful if it lasts 1 day?


How does a WSN intelligently manage it’s
limited power reserves?


Option 1: Reduce number of active
components on a node


Option 2: Put the entire node to sleep


All activity ceases


No routing capabilities


No sensing capabilities


Potential blind spot in the sensed area


Possible sub
-
graph disconnection


Intelligent Lighting Control


Routing


Data Analysis


Agent Environments:


Agilla, Mate, AFME


Characterised by the one
-
agent
-
per
-
node
approach


Weak notion of agency


Stack
-
based
approach


Messages sent using
lower layer


Massages received
from lower layer


Mediated
hibernation


A node is critical if


Connectivity
OR


Sensing are critical


Decision


persistence/timing


What if each layer can hibernate
independently?


Enforcing homogenous
timing policy can be highly
inefficient
-

experiments


Stack
-
based approach
allows passing messages
through hibernating layers


Solution: Allow each layer
to operate as an
autonomous agent.


Open source minimised footprint BDI agent
platform developed for resource constrained
devices.


Targets devices, such as mobile phones and
Sun SPOT leaf nodes.


Based on Agent Factory, a pre
-
existing agent
platform for desktop environments.


Conforms to the CLDC Java Specification.



AFME agents follow a sense
-
deliberate
-
act cycle.


In the control algorithm, initially perceptors are
fired and the belief set is updated. The desires are
then identified using resolution
-
based reasoning.
Various intentions are then chosen. Depending on
the nature of the intentions, various actuators are
fired.


AFME supports the Agent Factory Agent
Programming Language and augments it with an
infrastructure for resource bounded reasoning.



Perhaps the most obvious difference between
development for a desktop machine and a senor
concerns the limited spatiotemporal and energy
resources available.


This is coupled by the inherent uncertainty in
WSN domains.


What then does it mean to say an agent is
rational in circumstances where it does not have
the information or resources to determine the
course of action that yields maximum utility?


In this application, we are concerned with
altering sleep rates in a prudent manner to
improve system performance.


Should a system react quickly with a small
amount of data or continue operating as more
data is collected.


There is an inherent cost in controlling a system.


The macroscopic principle of uncertainty in
control theory.


The BDI model of agency acknowledges that
agents are resource bounded and will be
unable to achieve all of their desires even if
their desires are consistent.


An agent must fix upon a subset of desires an
commit resources to achieving them.


This subset is the agents intentions.


In essence, this is a classic 0
-
1 knapsack
problem.


5 metre node
separation


100m x 100m area
with a mobile
target


Active nodes
sample their
sensors every 10
seconds


% received


At present, the application has been
implemented using a stack based approach.


We have conducted experiments that
illustrate that when combined hibernation
strategies are adopted, it leads to poor
application performance.


Implement the agent based solution to the
problem using AFME.


Such an approach should improve
performance.


The problem: if we use a longer,
homogenous evaluation period the routing
component improves.


Need to break the homogenous evaluation
frequency while still allowing a node to
hibernate.


A MAS resident on a node could provide
such flexibility and power management.


More details may be found at:

http://www.prism.ucd.ie/index.html



AFME may be downloaded from:

http://sourceforge.net/projects/agentfactory