论文部分内容阅读
采用误差反传前向人工神经网络(Artificial neural network,ANN)建立了37种氯代芳烃的结构与其对孔雀鱼的急性毒性之间的定量关系模型(ANN模型)。以37种氯代芳烃的量子化学参数作为输入,急性毒性作为输出,采用内外双重验证的办法分析和检验所得模型的稳定性,所构建网络模型的相关系数为0.996 5、交叉检验相关系数为0.991 1、标准偏差为0.04、残差绝对值≤0.18,应用于外部预测集,外部预测集相关系数为0.988 4;而多元线性回归(Multiple linear regression,MLR)法模型的相关系数为0.949 6、交叉检验相关系数为0.928 8、标准偏差为0.14、残差绝对值≤0.32,外部预测集相关系数为0.950 5。结果表明,ANN模型获得了比MLR模型更好的拟合效果。
A quantitative relationship model (ANN model) was established between the structure of 37 chlorinated aromatic hydrocarbons and their acute toxicity to guppies using Artificial Neural Network (ANN). Taking 37 kinds of chlorinated aromatic hydrocarbons as input and the acute toxicity as output, the stability of the model was analyzed and verified by internal and external double verification. The correlation coefficient of the constructed network model was 0.996 5 and the cross-correlation coefficient was 0.991 1, the standard deviation is 0.04, the residual absolute value is ≤0.18, which is applied to the external prediction set, the correlation coefficient of the external prediction set is 0.988 4; and the correlation coefficient of the multiple linear regression (MLR) method is 0.949 6, The test correlation coefficient was 0.928 8, the standard deviation was 0.14, the residual absolute value was ≤0.32, and the correlation coefficient of the external prediction set was 0.950 5. The results show that ANN model has better fitting effect than MLR model.