CAJ | 학술논문

针对模拟退火（simulated annealing，SA）算法收敛速度慢，随机采样策略缺乏记忆能力，算法内在的串行性使其具有并行化问题依赖等缺点，提出了基于粒子群优化（particle swarm optimization，PSO）算法的并行模拟退火算法。该算法利用粒子群优化算法中个体的记忆功能引导算法在解空间中开展精细搜索，在反向学习算法基础上设计新的反向转动操作机制增加了算法的多样性，借助PSO的天然并行性克服了SA的并行问题依赖性，并在集群上实现了多Agent协同进化的改进算法。对Toy模型的蛋白质结构预测问题进行了仿真实验，结果表明该算法能有效提高求解问题的质量和效率。
침대모의퇴화（simulated annealing，SA）산법수렴속도만，수궤채양책략결핍기억능력，산법내재적천행성사기구유병행화문제의뢰등결점，제출료기우입자군우화（particle swarm optimization，PSO）산법적병행모의퇴화산법。해산법이용입자군우화산법중개체적기억공능인도산법재해공간중개전정세수색，재반향학습산법기출상설계신적반향전동조작궤제증가료산법적다양성，차조PSO적천연병행성극복료SA적병행문제의뢰성，병재집군상실현료다Agent협동진화적개진산법。대Toy모형적단백질결구예측문제진행료방진실험，결과표명해산법능유효제고구해문제적질량화효솔。
Traditional simulated annealing (SA) algorithm suffers from the problems of slow convergence, lack of memory in random sample and dependence on specific issues. This paper proposes a parallel SA algorithm based on particle swarm optimization (PSO) algorithm to solve the problems. By including the individual’s memory of PSO, the proposed algorithm can enhance the exploitation capability. In order to maintain diversity, a new opposition rota-tion learning strategy is introduced on the basis of opposition-based learning (OBL) algorithm. With the natural par-allelism of PSO, the shortcoming of problem-dependence of SA is solved. In addition, the proposed algorithm runs on cluster to achieve coevolutions. Experiments conducted with protein structure prediction based on Toy models show that the proposed algorithm outperforms both in convergence speed and solution quality.