Big Data Handling Of Nuclear Material From Parallel Meta-heuristic Algorithms

The coming of big data era is changing various aspects of our life, and also brings new opportunities and challenge to materials science. Recently, with the proposing of Materials Genome Initiative, the materials big data has been paid more and more attentions. This paper is exactly a research to one problem of materials big data which is the optimization of interatomic potential. Meta-heuristic algorithm is an efficient method to solve such problem,and genetic algorithms as well as particle swarm optimization are most popular in which.Recently, data storage of all fields including materials science are exploding and becoming more and more complex. As a result, the serial meta-heuristic algorithms cannot meet the needs yet, it is necessary to develop parallel meta-heuristic algorithms.This paper firstly analysis the intrinsic characteristics of the meta-heuristic algorithms,then parallelize the genetic algorithm and particle swarm optimization algorithm which are most famous in the meta-heuristic algorithms. Firstly, the Foster's design methodology is used to design the parallel genetic algorithms and particle swarm optimization, the fork/join strategy for genetic algorithm and fork/join combines concurrence strategy for particle swarm optimization are designed. Secondly, the OpenMP which is a shared-memory standard is used to parallelize the compute-intensive part of the two algorithms preliminary, it is proved that the efficiency promotes proportionally when the processors increases. Thirdly, MPI as well as OpenMP is used to parallelize the particle swarm optimization algorithm based on population divided. By this way, not only the efficiency, but also the performance of the algorithm is promoted. Furthermore, a particle swarm algorithm based on population divided and includes genetic mechanism and individual history information is proposed to improve the rate of convergence, it is proved has more advantage when the time complexity of problems are high than the algorithm without history information and genetic mechanism.Finally, the parallelized meta-heuristic algorithms are used to solve the problem of interatomic potential optimization with an example of irradiated SiC. The result proved that parallel meta-heuristic algorithms not only have less time complexity but also have excellent optimization property. Furthermore, the algorithm proposed by this paper includes individual history information and genetic mechanism is also proved performed better in such problems which are of high time complexity.