查看文章

Genetic Algorithms (GA’s) are of particular significance in applications with highly irregular search spaces, especially when the computing power requirement for exhaustive search is prohibitively large. GA’s are capable of finding near optimal solutions and highly amenable to parallelism and hence Distributed Genetic Algorithms (DGA’s) present an efficient solution to the time requirement issue. However, both the serial and the parallel implementations are subject to the genetic drift phenomenon, which drives the search into local optima if the exploration and exploitation are not well balanced. This work proposes a mechanism to achieve this balance by mimicking the natural catastrophes of life on earth; demes are partially destroyed in a randomized pattern and encouraged to reconstruct themselves towards better diversification and hence genetic drift avoidance or reduction.