CAJ | 학술논문

针对粒子群算法存在容易陷入局部最优的问题，提出一种改进杂交粒子群算法以强化全局搜索的能力。将粒子群优化与分布估计算法相结合，利用分布估计模型提取全局信息，用以指导种群大小的自适应变化。在全局信息变化显著时扩充种群，来加强算法在未知区域的探索；在种群大小达到上限时，利用全局信息更新较差个体，引导种群向最优区域集中；在全局信息或最优解均基本不变时，对种群进行重构，降低计算代价并防止陷入局部最优。7组标准函数的测试结果表明，改进算法优于其余几种与分布估计模型结合的杂交算法，在全部5组多模态函数的测试中其结果是最好的，其中在理论最小值未知的函数 F7上，所得最优值比其它算法提升了9.5～13.6。
침대입자군산법존재용역함입국부최우적문제，제출일충개진잡교입자군산법이강화전국수색적능력。장입자군우화여분포고계산법상결합，이용분포고계모형제취전국신식，용이지도충군대소적자괄응변화。재전국신식변화현저시확충충군，래가강산법재미지구역적탐색；재충군대소체도상한시，이용전국신식경신교차개체，인도충군향최우구역집중；재전국신식혹최우해균기본불변시，대충군진행중구，강저계산대개병방지함입국부최우。7조표준함수적측시결과표명，개진산법우우기여궤충여분포고계모형결합적잡교산법，재전부5조다모태함수적측시중기결과시최호적，기중재이론최소치미지적함수 F7상，소득최우치비기타산법제승료9.5～13.6。
For the easily falling into local optimum problem of the particle swarm optimization ,an advanced hybrid version was proposed to enhance the global searching ability .The particle swarm optimization was combined with the estimation of the distri-bution algorithm to obtain the global information ,and the information was used to guide the self-adaptation of the population size .The population was expanded to enhance the exploration in unknown regions when the global information varied obviously . If the population reached the upper limit size ,the worse particles were replaced so that all the particles moved to the promising regions .If the global information or the best value rarely changed ,the population was compressed and reconstructed to reduce the computational cost and to avoid falling into the local optimum .Results of experiments on seven benchmarks show that the proposed algorithm works better than other hybrid algorithms based on the estimation of the distribution model .Especially ,the proposed algorithm obtains the best results on five multimodal benchmarks ,and for the benchmark F7 with the unknown global optimum ,the best value is improved by 9.5-13.6 compared to others .