Bio-Inspired Computing For Complex And Dynamic Constrained Problems

Abstract

Bio-inspired algorithms are general-purpose optimisation methods that can find solutions with high qualities for complex problems. They are able to find these solutions with minimal knowledge of a search space. Bio-inspired algorithms (the design of which is inspired by nature) can easily adapt to changing environments. In this thesis, we contribute to the theoretical and empirical understanding of bioinspired algorithms, such as evolutionary algorithms and ant colony optimisation. We address complex problems as well as problems with dynamically changing constraints. Firstly, we review the most recent achievements in the theoretical analysis of dynamic optimisation via bio-inspired algorithms. We then continue our investigations in two major areas: static and dynamic combinatorial problems. To tackle static problems, we study the evolutionary algorithms that are enhanced by using a knowledge-based mutation approach in solving single- and multi-objective minimum spanning tree (MST) problems. Our results show that proper development of biased mutation can significantly improve the performance of evolutionary algorithms. Afterwards, we analyse the ability of single- and multi-objective algorithms to solve the packing while travelling (PWT) problem. This NP-hard problem is chosen to represent real-world multi-component problems. We outline the limitations of randomised local search in solving PWT and prove the advantage of using evolutionary algorithms. Our dynamic investigations begin with an empirical analysis of the ability of simple and advanced evolutionary algorithms to optimise the dynamic knapsack (KP) problem. We show that while optimising a population of solutions can speed up the ability of an algorithm to find optimal solutions after a dynamic change, it has the exact opposite effect in environments with high-frequency changes. Finally, we investigate the dynamic version of a more general problem known as the subset selection problem. We prove the inability of the adaptive greedy approach to maintain quality solutions in dynamic environments and illustrate the advantage of using evolutionary algorithms theoretically and practically.