| Summary: | The Bees Algorithm is a new population-based optimisation procedure which employs a combination of global exploratory and local exploitatory search. This thesis introduces an enhanced version of the Bees Algorithm which implements a fuzzy logic system for greedy selection of local search sites. The proposed fuzzy greedy selection system reduces the number of parameters needed to run the Bees Algorithm. The proposed algorithm has been applied to a number of benchmark function optimisation problems to demonstrate its robustness and self-organising ability. The Bees Algorithm in both its basic and enhanced forms has been used to optimise the parameters of a fuzzy logic controller. The purpose of the controller is to stabilise and balance an under-actuated two-link acrobatic robot (ACROBOT) in the upright position. Kalman filtering, as a fast convergence gradient-based optimisation method, is introduced as an alternative to random neighbourhood search to guide worker bees speedily towards the optima of local search sites. The proposed method has been used to tune membership functions for a fuzzy logic system. Finally, the fuzzy greedy selection system is enhanced by using multiple independent criteria to select local search sites. The enhanced fuzzy selection system has again been used with Kalman filtering to speed up the Bees Algorithm. The resulting algorithm has been applied to train a Radial Basis Function (RBF) neural network for wood defect identification. The results obtained show that the changes made to the Bees Algorithm in this research have significantly improved its performance. This is because these enhancements maintain the robust global search attribute of the Bees Algorithm and improve its local search procedure.
|
|---|
