Seeding the population
Bryant A. Julstrom · 1994
A hybrid genetic algorithm combines techniques particular to a problem with a genetic algorithm.Perhaps the simplest hybrid GA applies problem-specific information to seed the algorithm's initial population with one or more organisms known to be fairly good but then operates as a conventional GA.This paper describes the effect of such seeding on a genetic algorithm for the rectilinear Steiner problem, which seeks a shortest rectilinear Steiner tree on a set of given points.One version of the algorithm generated its initial population entirely randomly.The second seeded its population with one rectilinear Steiner tree derived from a minimal rectilinear spanning tree.The seeded algorithm identified shorter rectilinear Steiner trees more consistently and much more quickly than did the unseeded algorithm.It identified trees that are, on average, approximately 90% as long as minimal rectilinear spanning trees on the same points.This is shorter than trees found by other algorithms that approximate solutions to this problem.