Revisiting the Design of Adaptive Migration Schemes for Multipopulation Genetic Algorithms

Multipopulation Genetic Algorithms (MGAs) are island model genetic algorithms composed of spatially semi-isolated sub-populations, each evolving in parallel by its own pace and occasionally interacting with its neighborhoods by interchanging (usually good) individuals, called migration. Since the migration process is the kernel mechanism of MGAs for preventing premature convergence, many previous works have been devoted to the design of good migration schemes, including migration policy, migration interval, and migration rate, but very few work focusing on adaptive aspect of the migration schemes. In this study, we revisit this problem by inspecting the design of adaptive migration schemes from two perspectives, fitness-based, i.e., favoring the solution quality, or diversity-based, i.e., sustaining population diversity, and thereby we propose two new adaptive migration schemes, one is fitness-based and the other is diversity-based. A preliminary experiment on 0/1 knapsack problem shows that both of the new approaches are better than our previous methods, and the diversity-based approach is more effective than the fitness-based approach.