BP神经网络的IAPSOBPNN组合训练算法.pdfVIP

第22 卷 第9 期 重 庆 工 学 院 学 报( 自然科学) 2008 年9 月 Vol. 22 No. 9 Journal of Chongqing Institute of Te hnology(Natural S ien e) Sep. 2008 X BP IAPSOBPNN 黄 丽, 唐万梅 ( , 400047) : BP , BP IAP- SOBPNN( Parti le Swarm Optimization with Immunity Algorithm Ba k Propagation Neural Network) , BP . , , , , BP . : BP ; IAPSOBPNN ; ; :TP183 : A : 1671- 0924( 2008) 09- 0120- 07 IAPSOBPNN Combined Training Algorithms for BP Neural Network HUANG Li,TANG Wan-mei ( College of Mathemati s and Computer S ien es, Chongqing Normal University, Chongqing 400047, China) Abstract: BP neural network is easily trapped into lo al minima, and has low generalization ability. In terms of these problems, IAPSOBPNN( Parti le Swarm Optimization with Immunity Algorithm Ba k Propa- gation Neural Network) ombined training algorithm is proposed, namely: a ombination of immunity algo- rithm , parti le swarm optimization and ba k propagation algorithm. The ombined training algorithm is ap- plied to the nonlinear fun tion approximation and sto k pri e fore ast with omplex nonlinear dynami har- a teristi , and the simulation results indi ate that the algorithm avoids the network from lo al minima, en- han es the generalization ability of the network, provides a new idea for fixing the parameters of BP net- work. Key words: BP neural network; IAPSOBPNN ombined training algorithm; nonlinear fun tion approxima- tion; sto k pri e fore ast X : 2008- 03- 17 : ( KJ060818, KJ060804) ; ( 06XLB23) .