Pseudo Parallel Ant Colony Optimization for Continuous Functions

This paper presents a pseudo parallel ant colony optimization (ACO) algorithm in continuous domain. The variables of a solution are optimized by two parallel cooperative ACO-based processes, either of which attacks a relatively-independent sub-component of the original problem. Both processes contain tunable and untunable solution vectors. The best tunable vector migrates into the other process as an untunable vector through a migration controller, in which the migration strategy is synchronously sprung or adaptively controlled according to the temporal stagnation situation. Implementation of this mechanism is suitable for hardware which supports parallel computation, resulting in decline of unit computational cost and improvement of training speed. Optimization to a set of benchmark functions is carried out to prove the feasibility and efficiency of this parallel ACO algorithm.