摘要
针对BP算法存在的易陷入局部极小点、收敛速度慢、所设计的网络泛化能力不能保证等问题,提出了一种基于粒子群优化算法(PSO)的BP(PSO-BP)网络的权值调整新方法。该算法在基本BP算法的误差反传调整权值的基础上,再引入PSO算法的权值修正,从而建立了基于PSO-BP网络模型。基于此模型设计了规范手写体数字识别的分类器,采用随机手写数字样本进行了仿真实验,结果表明:PSO-BP算法提高了网络的稳定性,避免了BP算法容易进入平坦区、陷入局部极小等问题。
For BP algorithm has local minimum point easy, slow convergence and the generalization ability of network design can not be guaranteed and other issues, a particle swarm optimization （ PSO ） algorithm BP （ PSOBP）network is proposed, which can adjust the network value. On the basis of the basic BP algorithm backpropagation algorithm adjust weights, re-introduction PSO of the right to correction, which set up BP network model based on PSO algorithm. Handwritten numeral recognition classifier is designed based on this model, using random samples of the handwritten figure simulation results show that PSO-BP neural network algorithm enhances the stability of the network, avoids the BP algorithm easy access to fiat areas and a partial very small and so on.
出处
《传感器与微系统》
CSCD
北大核心
2009年第9期9-11,共3页
Transducer and Microsystem Technologies
关键词
粒子群优化
BP神经网络
数字识别
particle swarm optimization（PSO）
BP neural network
numeral recognition
