Parallel architecture design of genetic algorithms on HPC platform
Yan Liu, Wenxin Hu, Longzhe Han · Proceedings of the ACM Turing Celebration Conference - China · 2019
In the genetic algorithm for compute-intensive task, the execution time of fitness function increases rapidly. As a result, when the population size or evolution population increases, the convergence speed of the algorithm is very slow. Based on the supercomputer platform, this paper designs and implements a parallel genetic algorithm architecture with dynamic migration strategy. A hybrid parallel genetic algorithm based on "coarse-grained and master-slave" two-level parallel architecture is proposed. The proposed architecture is composed of hybrid parallel programming models. Compared with the traditional genetic algorithm implemented on the simple MPI architecture, the proposed structure achieves better performance. Experiments show that the convergence speed of the algorithm is significantly improved.