摘要
提出在BP网络的输入层与隐层或径基函数网络的输入层与隐层之间增加一个只有 2个结点的线性函数层(Z层 ) ,以构成基于BP网和基于径基函数网的降维映射网络。这两种网络均将多维输出Y (可为一维 )与多维输入X之间的非线性映射关系转变成与二维向量Z之间的非线性映射关系。网络学习后 ,就可以在由Z构成的二维映射平面上描绘出输出向量的等值线 ,通过这些等值线可全景式地、准确可靠地确定出样本数据集的最优操作区域 。
A liner function layer (Z layer) contained only two notes is inserted between the input and the hidden layers in BP network or in RBF network to form reducing dimensional mapping. By means of these two networks it can be achieved that the no n liner mapping between multi dimensional (Y) output (also one-dimensional) and input (X) converts into the one between Y and the two dimensional vector (Z). The network can produce the contours of output vectors (Y) in the Z plane from which the optimal operational region on the sample data set can be determined accurately and the purpose of concrete optimal design can be realized.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2001年第3期241-246,共6页
Computers and Applied Chemistry
关键词
神经网络
降维映射
混凝土
配合比
优化方法
优化设计
neural network, reducing dimensional mapping, mix proportion,optimal design of concrete
