Ant Colony Optimization with Parallel Subsolutions Heuristic

For solving NP-complete Travelling Salesman Problem (TSP) by combining solutions of its subproblems, we designed algorithm inspired by Ant Colony Optimization (ACO) modified by adding parallel subsolutions heuristic. ACO can he, parallelized simply simultaneous execution of the algorithm with eventual exchange of the best solutions between all computational units. This approach requires access to a whole state matrix (which is of size O(n(2))) for each of them. It can limit the size of solvable problems on special architectures with different, available memory capacities such as Cell Broadband Engine Architecture (CBEA). The presented algorithm keeps only one pheromone matrix in the memory of a main unit. The matrix is updated by subsolutions computed by ACO on other units in parallel. We show that this approach performs significantly better than greedy algorithm, even though it generates the whole solution froul solutions of subproblems.