| Summary: | One of the mam characteristics which set wireless sensor networks apart from traditional networks is the inclusion of energy consumption as the highest priority optimisation goal. This is because these types of networks work under the general concept that the system lifetime needs to be extended as much as possible, whilst at the same time achieving efficient data forwarding and preventing route disconnections due to sensor node failure. Hence, the use of energy efficient infrastructure such as clustering n:ay lengthen the lifetime of the network and prevent network connectivity degradation through the utilisation of cluster heads. Since the optimal selection of cluster heads in a network belongs to nondeterministic polynomial (NP) hard problem, the use of approximation algorithms such as Particle Swarm Optimisation (PSO) are generally more suitable due to its simplicity and outstanding search strength. This PhD thesis investigates the application of the PSO algorithm in clustering ofwireless sensor networks. In view of the need to prolong sensor network lifetime, a centralised, energy efficient, cluster-based protocol is developed using the PSO algorithm. A new cost function has been defined, which takes into consideration three important factors, namely the expected network energy consumption, the intracluster distance and the remaining energy of the cluster heads. The clustering problem is then transformed into an optimisation problem, and the PSO algorithm is employed to search for the optimal set of cluster heads. Simulation results demonstrate that the proposed protocol using PSO obtains better data delivery and network lifetime, as well as improves network connectivity over its comparatives. In addition, the results confirm the efficiency of PSG in clustering problems, compared to other evolutionary algorithms. This thesis also considers the use of the PSO algorithm in clustering the wireless sensor networks with mobile nodes. For this purpose, the 'mobility factor is taken into account when defining the cluster membership and selecting the cluster heads in order to maintain network connectivity. Simulation results prove that this approach outperforms other well known protocols in terms of data delivery and network lifetime. Finally, a dynamic multi-objective clustering algorithm which automatically determines the optimum number of clusters is introduced. This algorithm, based on binary PSG, eliminates the need to set the number of clusters a priori. Furthermore, the use of multi-objective PSG can tackle the difficulty of tuning the cost function weights that properly scales the sub-objectives. Performance evaluation through simulation exhibits the superior strength of this algorithm in enhancing the network survivability.
|
|---|
