摘要
针对过程神经网络在输入维数较高时存在时间代价过大的缺点,提出了基于核主元分析(KPCA)和离散Walsh变换的改进过程神经网络算法(IPNN-KPW)。该算法结合KPCA和离散Walsh正交基变换,减少了过程神经网络的输入计算代价;引入动量因子和自适应学习率,加速了网络收敛并有效地抑制了网络震荡。应用该算法对聚合反应中聚丙烯腈平均分子量建模,仿真实验结果验证了该算法的有效性,它能以较少的时间代价得到较高的模型精度。
Process neural network could handle the modeling problems of time related industry, but it needs a long time when the input dimension is high. in this paper, a new improved process neural network based on KPCA and Watsh （IPNN-KPW） is proposed. Both KPCA method and discrete Walsh transform are used to reduce the time cost of process neural network. Meanwhile, both momentum factor and self- adapting learning rate are introduced to accelerate the astringency of the network and keep down network＇s oscillation. IPNN-KPW is applied to model polyacrylonitrile （PAN） average molecular weight in polymerization, whose results verify the effectiveness of the proposed algorithm.
出处
《华东理工大学学报（自然科学版）》
CAS
CSCD
北大核心
2010年第4期585-590,共6页
Journal of East China University of Science and Technology
基金
国家自然科学基金项目(64974066)
