Feature-based selection of bio-inspired algorithms for constrained continuous optimisation

Abstract

Constrained continuous optimisation problems are widespread in the real-world and often very complex. Bio-inspired algorithms such as evolutionary algorithms (EAs) or particle swarm optimisation (PSO) algorithms have been successful in solving these problems. Recently, there has been an increasing interest in understanding the features of problems that make them hard to solve. These studies have been carried out for discrete and unconstrained continuous optimisation problems, to find the relationship between problem features and algorithm performance. To study the connection between algorithms and constrained optimisation problems (COPs), more practical perspectives of problem features analysis and their relations to algorithms are essential. Thus, this thesis contributes to the understanding of constrained optimisation problems and their constraint features that make them hard to solve by algorithms. We introduce an empirical feature-based analysis for COPs and bio-inspired algorithms. Furthermore, the relationships between the constraint features of given COPs and algorithms are studied here. By linking the features of the constraints and different bio-inspired algorithms, we design a new model for predicting the algorithm performance for COPs based on their constraint features. In this thesis, we present a novel approach to analyse constrained continuous optimisation problems based on their constraint features. Furthermore we use this knowledge to implement an automated feature-based algorithm selection model for constrained continuous optimisation.