Multi-objective particle swarm optimisation: methods and applications.
Al Moubayed, Noura
MetadataShow full item record
Solving real life optimisation problems is a challenging engineering venture. Since the early days of research on optimisation it was realised that many problems do not simply have one optimisation objective. This led to the development of multi-objective optimizers that try to look at the optimisation problem from di erent points of view and reach a set of compromised solutions among the di erent objectives. The presented research brings together recent advances in the eld of multi-objective optimisation and particle swarm optimisation raising several challenges. This is tackled from di erent aspects including the proposal of new archiving techniques to developing new methods and quality measures. Smart Multi-objective Particle Swarm Optimisation based on Decomposition (SDMOPSO) is rst proposed to incorporate multi-objective problem decomposition techniques with PSO. A novel archiving technique is developed using a clustering based mapping approach between the objective and solution spaces and is applied to general multi-objective optimizers. D2MOPSO is introduced as a new MOPSO that uses problem decomposition and a new archive utilising dominance based mapping between objective and solution spaces. Finally the thesis presents a novel multi-objective quality measure that uses mutual information to compare among solutions generated by di erent algorithms. The contributions are all tested on standard test suits and are used to solve two real-life problems: a) Channel selection for Brain-Computer Interfaces, and b) E ective cancer chemotherapy treatments. The two problems are real challenges in the two respective elds. Two di erent modelling approaches of the channel selection problem are presented: one is based on binary representation of the channels, while the other is continuous in a projected space of the channel locations. The results are very competitive with the commonly used methods.