The Original Bee Colony Algorithm proposed by Chong et al. (2006) models key features of the foraging cycle of a colony of honey bees. The Original Bee Colony Algorithm generally outperforms the Ant Colony Algorithm, but falls behind the Tabu Search Algorithm significantly. By modifying parts of the Original Bee Colony Algorithm to fit the proposed natural behavior of real honey bees, minor improvements are observed.
A novel algorithm inspired by the Original Bee Colony Algorithm was also introduced. The new algorithm incorporated the Bee Colony system onto a disjunctive graph representation of the job shop scheduling problem, where swaps are performed in the critical path of the current solution in each foraging cycle in the hope of locating the optimum solution. The initial version performed very poorly. Several modifications were implemented and the performance of the new algorithm steadily improved. The most significant improvement was observed when several dispatching rules were combined into the algorithm instead of one. Results further improved when ineffective dispatching rules were replaced with more effective ones. The latest results showed that performance of the novel approach is near to that of the Ant Colony Algorithm. The potential of the novel system lies in combining the characteristics of Swarm Intelligence with the proven effectiveness of Neighborhood Search.
Several future research areas were proposed, ranging from improving the efficiency of critical path calculation, to further enhancements to the neighborhood structure and search method, to visualization through an interactive Gantt chart system.
The results of this project seemed to suggest that a combination of various techniques usually leads to better performance. The author believes that the Bee Colony Optimization approach has potential yet to be revealed, especially for the combined method involving the Bee Colony Optimization and the Neighborhood Search. Further research into this area should prove to be fruitful.
