人工神经网络用于预测离子交换分配系数

The Prediction of Cation Exchange Distribution Coefficients with Artificial Neural Network

用前向网络和扩展的ｄｅｌｔａ－ｂａｒ－ｄｅｌｔａ算法对主族特征价阳离子的离子交换分配系数（Ｋ＿ｄ）进行了预测，对网络结构、学习次数进行了优化并研究了学习集的大小．１ｎＫ＿ｄ的均方根偏差（ＲＭＳ）小于７％．

The ion exchange distribution coefficients of cat ions having characteristie valence in main groups were predicted by using extented delta-bar-delta neural networks, The effect of structure of network, the learning Epochs and the size of learning set on predicted results was investigated.The suitable conditions are:input nodes 4; output node: 1; the learning epochs: 3 000; the size of learning set:112. The RMS(root-mean-square error) is smaller than 7%under the suitable conditios.