多氯酚定量构效关系人工神经网络信息流分析

Analyzing the information flow of chlorophenol QSAR artificial neural network model

针对目前有机污染物定量构效关系(QSAR)人工神经网络(ANN)模型构建中存在的问题,在多氯酚QSAR-ANN模型数据研究基础上,提出了人工神经网络信息流分析方法.通过比较多氯酚定量构效关系人工神经网络模型中的连接权值与阈值,绘制人工神经网络模型信息流流向分布图,可直观地显示网络内部信息的传播途径,找出影响输出层信息的隐含层的主要节点,进而由隐含层主要节点确定输入层主要节点,达到对人工神经网络模型输入节点的筛选.

Aimed at the problem in establishing QSAR-ANN model of organic pollutant at present, the analysis technique of ANN information flow was suggested, based on the study of chlorophenol QSAR-ANN model data. The distribution figure of ANN model information flow direction was made through comparing the connection value and the threshold in the chlorophenol QSAR-ANN model. The figure could show clearly the propagation way of ANN internal information for finding the key nodes of hidden layer that effect the information of output layer, and further define the key nodes of input layer, through nodes of hidden layer, so as to achieve the screening of input nodes.