DELTA势阱改进QPSO优化BP算法及其应用  

BP neural network optimized with QPSO algorithm improved by DELTA potential trough and its application

作　　者：于凤玲[1] 周扬[2] 陈建宏[3] 周汉陵[3] 

机构地区：[1]五邑大学经济管理学院,广东江门529020 [2]湖南文理学院资源环境与旅游学院,湖南常德415000 [3]中南大学资源与安全工程学院,长沙410083

出　　处：《中国科学院大学学报（中英文）》2015年第3期416-421,共6页Journal of University of Chinese Academy of Sciences

基　　金：国家自然科学基金(51374242)资助

摘　　要：为了改进BP算法预测性能,提出QPSO-BP模型.该模型采用DELTA势阱改进的量子粒子群(QPSO)算法优化BP网络的权值与阈值,然后利用各年的GDP数据进行训练和预测.结果表明:经过DELTA势阱改进的QPSO优化BP算法模型比PSO-BP模型和BP神经网络更稳定,预测精度更高且泛化能力更强.与文献中所用模型的运算结果相比较,这种改进模型运算结果的相对误差和平均误差更小,在准确性上也有一定的优势.To improve the generalization ability of BP network for prediction, a BP neural network optimized with QPSO is proposed. This model uses the QPSO improved by δ potential trough to optimize the initial values of weights and thresholds of BP network. Then the data of each year＇s GDP are selected in training and prediction. The experiments show that the QPSO-BP network optimized by using δ potential trough produces stable prediction results. Compared with the prediction models of PSO-BP and BP, the proposed model has a better generalization ability and a higher accuracy. In addition, the calculation results of the improved QPSO-BP optimization algorithm model have smaller relative errors and average errors compared with the results of themodels in the literature.

关 键 词：BP神经网络 PSO模型 QPSO模型 δ势阱 GDP 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] F201[自动化与计算机技术—控制科学与工程]