用改进的遗传算法训练神经网络以研究有机磷杀虫剂的构效关系

QSAR STUDY OF ORGANIC PHOSPHATE INSECTICIDE WITH IMPROVED GENETIC ALGORITHM CHEN Feng (Department of Chemical Engineering, Zhejiang University, Hangzhou,310027)

结合改进的遗传算法（ＩＧＡ）和误差反传算法（ＢＰＡ）训练人工神经元网络，用以对三元不对称有机磷构效关系进行研究。作者保留了ＢＰＡ作为权值基本训练方法简捷的优点，又利用遗传算法的全局搜索性，以克服ＢＰＡ陷入局部极小点的缺陷，使两者集成在一起，通过因子α调整遗传算法和误差反传算法的结合程度，发挥各自的长处，达到训练过程的优化。通过实例表明建立了更准确的数学模型，具有较好的实用价值。

QSAR of a symmetric organic phosphate insecticide was studied by combining the advantage of improved genetic algorithm(IGA) and the error back propagation algorithm (BPA) for effectively training multilayer artificial neural network (MAN). The advantage in simplicity of BPA was employed as basic method, and the nature of global optimization by GA is utilized to overcome the weakness that the conventional BPA is easy to fall in local minima. Through a procedure of adjusting the combination degree of IGA and BPA with a factor α, the proposed approach has been proved to obtain the final mathematical model more accurate and applicable to practical usages.