Bees Algorithm : theory, improvements and applications

In this thesis, a new population-based search algorithm called the Bees Algorithm (BA) is presented. The algorithm mimics the food foraging behaviour of swarms of honey bees. In its basic version, the algorithm performs a kind of neighbourhood search combined with random search and can be used for both combinatorial and functional optimisation. In the context of this thesis both domains are considered. Following a description of the algorithm, the thesis gives the results obtained for a number of complex problems demonstrating the efficiency and robustness of the new algorithm. Enhancements of the Bees Algorithm are also presented. Several additional features are considered to improve the efficiency of the algorithm. Dynamic recruitment, proportional shrinking and site abandonment strategies are presented. An additional feature is an evaluation of several different functions and of the performance of the algorithm compared with some other well-known algorithms, including genetic algorithms and simulated annealing. The Bees Algorithm can be applied to many complex optimisations problems including multi-layer perceptrons, neural networks training for statistical process control and the identification of wood defects in wood veneer sheets. Also, the algorithm can be used to design 2D electronic recursive filters, to show its potential in electronics applications. A new structure is proposed so that the algorithm can work in combinatorial domains. In addition, several applications are presented to show the robustness of the algorithm in various conditions. Also, some minor modifications are proposed for representations of the problems since it was originally developed for continuous domains. In the final part, a new algorithm is introduced as a successor to the original algorithm. A new neighbourhood structure called Gaussian patch is proposed to reduce the complexity of the algorithm as well as increasing its efficiency. The performance of the algorithm is tested by use on several multi-model complex optimisation problems and this is compared to the performance of some well-known algorithms.