Generating optimised partitions for parallel finite element computations employing float-encoded genetic algorithms

摘要

This paper presents an algorithm for automatic partitioning of unstructured meshes for parallel finite element computations employing float-encoded genetic algorithms (FEGA). The problem of mesh partitioning is represented in such a way that the number of variables considered in the genome (chromosome) construction is constant irrespective of the size of the problem. In order to accelerate the computational process, several acceleration techniques like constraining the search space, local improvement after initial global partitioning have been attempted. Finally, micro float-encoded genetic algorithms have been developed to accelerate the computational process. Numerical experiments have been conducted to demonstrate the effectiveness of the GA based partitioning algorithms. The proposed algorithms have been tested on unstructured meshes describing practical engineering problems. Apart from these meshes, several benchmark problems available in the literature are also considered to evaluate the performance of the GA based partitioning algorithms. Results indicate that the proposed algorithms are qualitatively superior to popular spectral approaches and multilevel algorithms. It was also shown through numerical experiments that the micro-float-encoded genetic algorithms provide faster solutions with slight reduction in the quality when compared to float-encoded genetic algorithms.