人工神经网络输入层节点筛选规则的确定

针对目前应用人工神经网络构建定量构效关系模型中输入层节点筛选存在的问题，提出了采用人工神经网络对网络输入层节点进行筛选，归纳出筛选规则，利用此规则可简便、快速地对多氯酚生物毒性预测人工神经网络模型的输入层节点进行筛选，输入层节点由最初的24个筛选到最后的3个。对筛选过程中不同输入层节点构建的网络模型质量和预测能力进行比较，得出含有较少输入层节点的人工神经网络模型的预测能力较高，运算速度较快，该规则的建立有利于进一步开展有机化学品生物毒理学的研究，并且该方法可以推广到人工神经网络应用的其他领域。

Study on the Screening Rule of Input Nodes in Artificial Neural Network

A new screening rule of input nodes in Artificial Neural Networks (ANN) is given. The new screening rule overcomes the problem, which some important nonlinear information between input nodes and output nodes may be omitted as screening input nodes in Quantitative Structure Activity Relationships (QSARs) models applyingANN.The structure parameter in multi chlorophenol QSAR model can be screened fastly and simply by using it. The results show that the structure parameters are cut down from 24 to 3, and the model quality and prediction ability in ANN are not reduced as the number of nodes in input layer is cut down, but are improved. In addition, the method expedites the convergence of networks model. So the method establishes the foundation for further developing the mechanism research of the toxicity of organic chemicals on biology and may be popularized in the other fields using ANN.