预防过拟合现象的 人工神经网络训练策略及其应用

长江科学院院报 ›› 2002, Vol. 19 ›› Issue (3): 59-61.

预防过拟合现象的人工神经网络训练策略及其应用

覃光华, 丁晶, 陈彬兵

出版日期:2002-06-01 发布日期:2012-03-05

Learning strategies of artificial neural networks for preventing over-training and their application

QIN Guang-Hua, DING Jing, CHEN Bin-Bing-

Online:2002-06-01 Published:2012-03-05

摘要/Abstract

摘要： 针对人工神经网络技术在水文研究中出现的过拟合现象，从训练样本的构成入手，提出了两种新的训练策略，即择优检验和加权检验。前者通过计算比较，选择最优的训练样本；后者对不同训练样本的评定结果进行加权，以得到最终的结果。应用实例表明，此两种方法能有效地减轻神经网络中过拟合的缺点。

Abstract: Aiming at the problem of over training in hydrologic research using artificial neural networks (ANNs) and considering the composition of training samples,two new learning strategies were presented,i.e,optimized detecting method and weighted detecting method.The former chooses the optimal training data after comparing the validation sets.The latter endues the different sets with different weights in order to get a final validation outcome.The application of the two new methods at Zipingpu Hydrographic Station on Minjiang River upstream indicates that they can resolve the problem of over-training effectively.

覃光华, 丁晶, 陈彬兵 . 预防过拟合现象的人工神经网络训练策略及其应用[J]. 长江科学院院报, 2002, 19(3): 59-61.

QIN Guang-Hua, DING Jing, CHEN Bin-Bing- . Learning strategies of artificial neural networks for preventing over-training and their application[J]. JOURNAL OF YANGTZE RIVER SCIENTIFIC RESEARCH INSTI, 2002, 19(3): 59-61.