摘要
在噪声小的ＰＬＳ（偏最小二乘法）空间上，样本集的局部投影可被用作ＢＰＮ（反向传播网络）的输入元素以建立一种“平衡”的神经网络结构，这种结构在很大程度上克服了通常ＢＰＮ过拟合的缺点。在ＰＬＳ子空间优化区，利用非线性逆映照技术设计的基于期望目标值的样本可通过ＰＬＳ－ＰＮ方法预报和选取。本文还利用此方法设计了若干以初始容量为目标的Ｎｉ／ＭＨ电池阴极材料。
The partial projection of sample set with less noise in Partial Least Square(PLS) space is used as input elements of Back Propagation Network (BPN)to build a balance neural network structure which can overcome the shortcoming of overfitting in BPN in a great extent.In the optimal region of PLS sub-space,the samples,which are based on the predicted target values and designed by the non-linear inverse mapping technique,can be predicted and selected using PLS-BPN method, The PLS-BPN method is applied in design of the cathode material of the Ni/MH battery.
出处
《计算机与应用化学》
CAS
CSCD
1996年第4期253-256,共4页
Computers and Applied Chemistry
基金
国家自然科学基金
"863"资助
