摘要： 利用生物发酵过程运行数据为样本，对BP神经网络进行训练，考察了网络结构（隐含层数、隐含结点数议及学习率η和动量参数α对收敛精度与收敛速度的影响，提出了稳定区的概念和多输出网络分割的思想，并给出了得到较好收敛速度与精度的学习方案。
关键词:
人工神经元网络,
反向传播算法,
稳定区,
多输出网络分割
Abstract: Artificial Neural Network (ANN) was trained using biological fermentation data as samples.The effects of ANN structure(Hidden Layers Numbers & Units number in each Hidden Layer), learn rate ηand momentum constant α，on learn precision and speed were studied in details.We proposed the conception of stable area & multi-output network division, and provided optimized methods of Back-propagation (BP) training.
中图分类号:
张明德,冯闻铮,董敏,刘福祯,乔长阁. 改进的BP神经网络收敛性的实验研究[J]. , 1998, 24(05): 27-29.
Zhang Mingde, Feng wenzheng, Dong Min, Liu Fuzhen, Qiao Changge. Study on Convergence Property of Progressed Back-propagation Neural Network[J]. , 1998, 24(05): 27-29.