论文部分内容阅读
本文利用误差反向传播(BP)的人工神经网络(ANN)模型研究了对位取代苯酚衍生物的生物活性与其结构及物理化学性质参数之间的定量构效关系。优化了ANN模型的参数设计,提出了动态调节网络学习速率的经验规则以改善网络的性能。采用f(x)=1/(1+e~(-x))作为网络节点的输入输出转换函数的三层神经网络具有较佳性能,当取隐含节点数为10时,该网络预测26个对位取代苯酚衍生物生物活性的均方误差(mse)为0.036,优于常规构效关系预测结果。
In this paper, the artificial inverse neural network (ANN) model of error backpropagation (BP) was used to study the quantitative structure-activity relationship between the biological activities of p-substituted phenol derivatives and their structural and physico-chemical properties. The parameter design of ANN model is optimized, and the rule of experience for dynamically adjusting network learning rate is proposed to improve the performance of the network. The three-layer neural network using f (x) = 1 / (1 + e ~ (-x)) as the input-output transfer function of the network node has better performance. When the number of hidden nodes is 10, The mean square error (mse) of the para-substituted phenol derivatives was 0.036, which was better than the predicted results of the conventional structure-activity relationship.