Optimization of large electric power distribution using a parallel genetic algorithm with dandelion strategy

Abstract

The study of electrical distribution of primary networks design is oriented to reduce the construction costsand the energy losses by transmission. The topology for the implementation of distribution networks may vary accordingto the geographical characteristics of the final users and requires specialized optimization solutions with metaheuristicsto improve the energy performance of the electrical power systems. A parallel genetic algorithm (PGA) is proposedto optimize a tree-based topology for large-scale electric power distribution networks. The proposed PGA uses thedandelion code, which allows obtaining tree-feasible solutions within each iteration of the PGA. This cannot be achievedwith other metaheuristic approaches directly. Eight cores are used simultaneously. We achieve a 22.05% improvementwhen compared to the tree-feasible solutions obtained with its sequential version. Moreover, the computational timerequired by the PGA is on average 23 times lower than the sequential version. Finally, we find feasible solutions forinstances of the problem with up to 50,000 nodes