摘要
通过类比信息传递过程中的一般测不准关系式,研究BP网络出现过拟合现象时,代表网络学习能力的权值改变量与反映网络推广能力的未参与训练的检验样本的辨识误差之间的关系.在该关系式中引入一个过拟合参数P,通过对岷江流域紫坪埔站的流量预报,说明在判断网络是否出现过拟合现象时,计算过拟合参数P是一种有研究意义的参考.
Compared with a general uncertainty relation in the information theory the paper researches the relation between the change of weight values which represent learning ability and the discrimination error of unlearning sample sets which represents generalization ability. An over-fitting parameter P is introduced; and the uncertainty relation is utilized for predicting river flow of Mingjiang River at Zipingpu station. The results show that the over fitting parameter P is significative to the problem of over-fitting.
出处
《武汉大学学报（工学版）》
CAS
CSCD
北大核心
2006年第6期55-58,共4页
Engineering Journal of Wuhan University
基金
四川大学青年科学研究基金资助
关键词
神经网络
过拟合
流量预报
neural networks
over-fitting
flow forecasting
