首页 | 本学科首页   官方微博 | 高级检索  
     

ACO—BP在神经网络训练中的研究与应用
引用本文:王鸽,蒲蓬勃.ACO—BP在神经网络训练中的研究与应用[J].计算机仿真,2009,26(12):136-140.
作者姓名:王鸽  蒲蓬勃
作者单位:山东科技大学信息工程系,山东,泰安,271021
基金项目:山东科技大学科学研究"春蕾计划"项目 
摘    要:针对神经网络收敛速度慢、易于陷入局部最优等问题,可将蚁群算法与人工神经网络相融合的方法来解决,但容易出现训练时间与训练精度、泛化能力之间的矛盾.为解决上述矛盾,提出将蚁群优化算法与反向传播算法相融合共同完成神经网络训练的方法.算法首先采用蚁群优化算法对网络权值进行整体寻优,克服反向传播算法容易陷入局部最优的不足再以找到的较优的权值为初值,采用反向传播算法做进一步的寻优,克服单一训练网络时间较长、精度不高的缺点.最后对ACO-BP与反向传播算法进行了比较,给出两种算法在不同隐结点数目下的检验误差值和两种网络在矿选指标中的应用效果.通过对实验结果的分析.表明ACO-BP算法要优于反向传播算法.

关 键 词:蚁群优化  反向传播  神经网络

Application of ACO-BP in Neural Network Training
WANG Ge,PU Peng-bo.Application of ACO-BP in Neural Network Training[J].Computer Simulation,2009,26(12):136-140.
Authors:WANG Ge  PU Peng-bo
Abstract:The integration of ant colony optimization algorithm with artificial neural network algorithm can solve some problems,such as slow convergence speed and easily falling into local optimum. But there is a conflict among training time, training accuracy and generalization ability. To solve the conflict, the paper proposed a new neural net-work training algorithm, named ACO-BP algorithm. ACO-BP scheme adopts ACO to search the optimal combina-tions of weights in the solution space. It overcomes the lack of a local optimum in BP algorithm. Then it uses BP algo-rithm to obtain accurate optimal solutions quickly, which overcomes the lack of single training network. At last, the paper compares ACO-BP algorithm with BP algorithm. It gives the results of test error in different hidden nodes and the application effect in mine selection index. The experiment results show that the ACO-BP model can obtain better predictive results than BP model.
Keywords:Ant colony optimization(ACO)  Back propagation(BP)  Neural network
本文献已被 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号