摘要 应用改进的BP网络法进行部分硝基芳烃的QSAR研究，用所建立的模型进行毒性预报，并与传统的BP网络和多元线性回归模型比较，表明改进的BP网络模拟和预报能力均优于传统的BP网络和多元线性回归模型。在学习集中，改进的BP网络，传统的BP网络，多元线性回归三者的均方差分别为:0.0240,0.010 7, 0.044 7,决定系数为:0.936 8,0.971 8,0.882 4;预报集中，三者的均方差分别为:0.044 0,0.055 4,0.077 2,决定系数为:0.8216,0.775 3,0.687 0。文中还讨论了网络改进的理由，以及网络结构中某些参数的选定问题。

    中图法分类号 R181.12, R994.6

Abstract Three kinds of quantitativestructure-activity relationship (QSAR) models for nitrobenzene derivatives wereconstructed by using the improved BP network and multivariate linear regression analysisand used to predict toxicities of the nitrobenzene derivatives not included in thetraining set. By comparing the calculated values and the experimental results, it wasconcluded that BP network approach for QSAR study of environmental pollutants was betterthan traditional BP network and multivariate linear regression analysis. For above threemodels, in learning set, MSE was 0.024 0 ......