Particle Swarm Optimisation Utilising Adaptive Fitness Function Generation

Adam Klyne

Abstract


The contribution of this paper is an algorithm for integrating
an anomaly detector with particle swarm optimisation.
We do so in order to achieve task allocation of resources when
the signature of a task is not well understood in advance. For our
proposed system we first discuss two candidate anomaly detection
approaches; a single layer feed forward neural network and
background subtraction. Once tasks are identified we then allow
a swarm intelligence system to autonomously allocate resources
to them. We found that both techniques work well for detecting
scene anomalies, with the neural network being more tolerant to
minor aberrations in the input data. In situations where the input
is tightly controlled we found that the background subtraction
method provides superior performance.