基于种子的多种群多策略并行PSO用于原子簇结构优化

SEED-BASED MULTI-GROUP MULTI-STRATEGY PARALLEL PARTICLE SWARM OPTIMIZATION STRUCTURAL OPTIMIZATION FOR ATOMIC CLUSTERS

针对Lennard-Jones(LJ)原子簇的结构优化问题,提出一种基于种子的多种群多策略并行PSO(SS-MG-MS-PPSO)。该算法使用种子策略(SS)初始化种群,再将初始化的种群划分为具有不同搜索策略(RDPSO和FC-RDPSO)的子种群独立并行进化,通过发挥不同搜索策略的优势从而找到原子簇的最优结构或近似最优结构。实验结果表明,SS-MG-MS-PPSO对于原子数在2到35之间的原子簇搜索到其最优或近似最优结构,且相比其他算法更为稳定、高效,为原子簇的结构优化提供了一种新的解决思路。

Aiming at the structural optimization problem of Lennard-Jones(LJ) clusters, a seed-based multi-group multi-strategy parallel PSO algorithm(SS-MG-MS-PPSO) is proposed. The algorithm used the seed strategy(SS) to initialize the population, then divided the initialized population into independent parallel evolution of sub-populations with different search strategies(RDPSO and FC-RDPSO), and found the optimal structure or approximate optimal structure of atomic clusters by exploiting the advantages of different search strategies. The experimental results show that the SS-MG-MS-PPSO algorithm searches for the optimal or approximate optimal structure for the atomic clusters with atomic numbers between 2 and 35, and is more stable and efficient than other algorithms, as well as provides a new solution for the structural optimization for atomic cluster.