基于粒子群优化算法的神经网络在农药定量构效关系建模中的应用  

Application of Neural Networks Based on Particle Swarm Algorithm for Modeling Quantitative Structure-Activity Relationships of Herbicides

作　　者：张丽平[1] 俞欢军[1] 陈德钊[1] 胡上序[1] 

机构地区：[1]浙江大学化工系,杭州310027

出　　处：《分析化学》2004年第12期1590-1594,共5页Chinese Journal of Analytical Chemistry

基　　金：国家自然科学基金资助项目(No.20276063)

摘　　要：神经网络模型能有效模拟非线性输入输出关系,但其常规训练算法为BP或其它梯度算法,导致训练时间较长且易陷入局部极小点。本实验探讨用粒子群优化算法训练神经网络,并应用到苯乙酰胺类农药的定量构效关系建模中,对未知化合物的活性进行预测来指导新药的设计和合成。仿真结果表明,粒子群优化算法训练的神经网络不仅收敛速度明显加快,而且其预报精度也得到了较大的提高。Neural networks (NNs) have become one of idea tools in modeling nonlinear relationship between inputs and desired outputs. However, the training of NNs by conventional back-propagation method, i.e. the BP-NNs, has intrinsic vulnerable weakness in slow convergence and local minima. In this work, particle swarm optimization (PSO) was proposed to train NNs, named PSO-NNs, to establish a quantitative structure-activity relationships model for herbicidal N-(1-methyl-l-phenylethyl) phenylacetamides, which provides a rational basis for understanding mechanisms of biological performance and how to alter chemical structure to achieve improved performance. Since neural networks can express complicated structure-activity relationships, and the combination of PSO Algorithms can help the networks to jump out from the local optimal points. The numerical results showed that PSO-NNs exhibit faster convergence rate and better predicting accuracy than the BP-NNs.

关 键 词：神经网络模型 粒子群优化算法 建模 局部极小 仿真结果 梯度算法 收敛速度 新药 乙酰胺 定量构效关系 

分 类 号：TQ450[化学工程—农药化工]