基于多粒子信息共享策略的PSO 小波网络模型.pdfVIP

35 5 () http: / /www. kustjourna.l com/ V ol 35 N o5 20 10 10 Journal ofK unm ing U n iversity of Sc ience and T echnology ( Sc ience and T echnology) Oct 2010 do:i 10 3969 / j issn 1007- 855x 2010 05 012 PSO 1 2 2 2 甘旭升, 端木京顺, 丛 伟 , 赵录峰 ( 1, 7 10123; 2 , 7 10038) : 针对粒子群优化算法在训练小波网络存在的早熟收敛问题, 提出一种改进的粒子群优化 算法 该算法 过引入多粒子信息共享策略, 使种群中各粒子共享多 个粒子的有用信息, 以期增 加种群多样性, 减少各粒子在进化早期被吸引到最优粒子附近的可能性, 提高小波网络的建模质 量 仿真表明, 相对于 BP 算法和标准粒子群优化算法, 本文算法在训练小波网络方面估计精度 更高, 收敛速度更快, 并有效解决了早熟收敛和局部最优问题 : 粒子群优化; 小波网络; 信息共享; 早熟收敛 : TP 181 : A : 1007- 855X ( 2010) 05 - 0052 - 04 PSOW aveletNeuralNetworkModel ased onMulti-Particles Information Sharing Strategy 1 2 2 2 GA N X u-sheng , DUAN MU J ing-shun , CONG W e i, Z HAO Lu-feng ( 1 X ijing U n iv ers ity, X ian 7 10 123, China; 2 Eng ineering College, A ir F orce Eng ineer ing U n iversity, X ian 7 100 38, Ch ina) Abstract: For prem aturity convergence in apply ing P artic le Sw arm Opt mi ization ( PSO ) algorithm to training of W ave let N eural N etw ork (W NN) , an mi proved PSO algorithm is proposed T o am e liorate popu la tion d iversity, a m ulti-p art icles in form at ion sh aring strategy is introdu ced into S mi p le PSO a lgorithm in order to redu ce the like l-i hood that each p art icle in population is attracted to the neighborhood o f the optmi um particle during the early pe- riod o f evolution S mi ulat ion s show th at, by contrast w ith BP and S mi p le PSO algorithm, th is a lgorithm is better in accuracy, convergence rate and prem aturity elmi inat ion in W NN tra in ing Key words: particle sw arm optmi ization;