Design and implementation of bio-inspired heterogeneous particle swarm optimisation algorithms for unconstrained and constrained problems

This thesis presents various novel non-bio and bio-inspired heterogenous PSO algorithms as general-purpose optimisers to handle real-valued unconstrained and constrained real-world problems. The proposed algorithms address various limitations of the canonical PSO, such as rapid loss of population diversity, which is closely associated with the problem of premature convergence to local optima. In general, the proposed algorithms utilise various mechanisms to improve the overall performance of the PSO algorithm by enabling particles to learn from distinct exemplars, maintaining better control of the flow of information between agents, through diversity-aware self-organised subswarms, design and reorganisation of topological structures, manipulation of dynamics of the particle population, and hybridisation. The algorithms presented in this thesis are global, population-based, single-solution, single-objective, non-derivative algorithms that can cope with real-valued unconstrained and constrained optimisation problems irrespective of features contained in the given problem. The algorithms perform well in both high and relatively low-dimensional search spaces. The experimental results obtained on three well-recognised CEC benchmark test suites at various dimensions, and 29 non-convex constrained real-world problems, indicate that the algorithms are highly competitive candidates as general-purpose optimisers. Detailed analysis of the results and of the search dynamics are presented.