Computational Intelligence Algorithms for Optimisation of Wireless Sensor Networks

Recent studies have tended towards incorporating Computation Intelligence,
which is a large umbrella for all Machine Learning and Metaheuristic
approaches into wireless sensor network (WSN) applications
for enhanced and intuitive performance. Meta-heuristic optimisation
techniques are used for solving several WSN issues such as energy
minimisation, coverage, routing, scheduling and so on. This research
designs and develops highly intelligent WSNs that can provide the
core requirement of energy efficiency and reliability. To meet these
requirements, two major decisions were carried out at the sink node
or base station. The first decision involves the use of supervised and
unsupervised machine learning algorithms to achieve an accurate decision
at the sink node. This thesis presents a new hybrid approach
for event (fire) detection system using k-means clustering on aggregated
fire data to form two class labels (fire and non-fire). The resulting
data outputs are trained and tested by the Feed Forward Neural
Network, Naive Bayes, and Decision Trees classifier. This hybrid approach
was found to significantly improve fire detection performance
against the use of only the classifiers. The second decision employs
a metaheuristic approach to optimise the solution of WSNs clustering
problem. Two metaheuristic-based protocols namely the Dynamic
Local Search Algorithm for Clustering Hierarchy (DLSACH) and Heuristics
Algorithm for Clustering Hierarchy (HACH) are proposed to achieve
an evenly balanced energy and minimise the net residual energy of
each sensor nodes. This thesis proved that the two protocols outperforms
state-of-the-art protocols such as LEACH, TCAC and SEECH
in terms of network lifetime and maintains a favourable performance
even under different energy heterogeneity settings.