论文部分内容阅读
采用误差反向传播学习(BP)的一个双层神经网络,以分子结构中不同基团作为描述码,预估芳香族多硝基化合物的生成焓,研究了网络参数及分子结构描述码的影响,同时用分子子图法进行了多元线性回归,取得了较好的结果(其回归方程相关系数达到0.9967),计算结果的绝大多数相对误差在10%范围以内。
Using a BP neural network with error backpropagation learning (BP), the formation enthalpies of aromatic polynitro compounds are predicted by using different groups in the molecular structure as the descriptive code. The effects of network parameters and molecular structure descriptors , While using molecular subgraph method for multiple linear regression, and achieved good results (regression equation correlation coefficient of 0.9967), most of the results of the calculation of the relative error within 10%.