Adaptive search heuristics applied to numerical optimisation

Penev, Kalin. (2004). Adaptive search heuristics applied to numerical optimisation. PhD thesis, Nottingham Trent University.

Abstract

The objective of the research project involves investigation of evolutionary computational methods, in particular analysis of population-based search heuristics, and abstraction of core cognition, which may lead to the design of a novel adaptive search algorithm capable of high performance and reliability. The thesis proposes a novel adaptive heuristic method called Free Search (FS). Free Search can be classified as a population-based evolutionary computational method. It gradually changes a set of solutions until satisfaction of certain criteria. The algorithm operates with real-value numbers. It is designed for continuous or partially discontinuous search space. Free Search harmonizes several advanced ideas, which lead to high overall performance. The study includes exploration of selected population-based evolutionary methods, namely : real-value coded Generic Algorithm BLX-a; Particle Swarm Optimisation (PSO); Ant Colony Optimisation (ACO); and Differential Evolution (DE). Common, substantial for the search purposes features, relationships and events are abstracted from the algorithms analysed. The events abstracted are generalised in a theoretical model of population-based heuristic search. The model supports significantly identification of essential advantages and disadvantages of population-based search algorithms and leads to establishment of a novel concept different from other evolutionary algorithms. Free Search together with GA BLX-a, PSO and DE are applied to heterogenous, numerical, non-linear, non-discrete, optimisation problems. The results are presented and discussed. A comparative analysis demonstrates better overall performance of FS than other explored methods. The capability of FS to cope with all tests illustrates a new quality - adaptation to the problem without concrete or specialised configuration of the search parameters. Free Search is tested aditionally with a hard, non-linear, constrained optimisation problem - the so-called bump problem. FS outperforms other methods applied to that problem. Results achieved from Free Search, currently unapproachable for other search algorithms are presented. Free Search opens a new area for research in the domain of adaptive intelligent systems. It can contribute also in general to Computer Science in the modelling of uncertain individual behaviour.

Actions (login required)

View Item