基于神经网络的双指标正交试验设计与分析

Two indexes of design and analysis of orthogonal experiments about a novel phosphorus free builder

摘要：以丙烯酸和马来酸酐为原料合成了一种新型无磷洗涤助剂MA-co-AA，以L16(4^5)正交实验的结果，采用误差反向传播EBP(Error Back Propagation)算法，完成了神经网络的学习，据此考察了单体配比、引单质量比、反应温度、反应时间、单体滴加时间对无磷助洗剂的螯合、分散双性能指标的影响。在给定误差范围内，预测了具有最佳性能产品的合成条件。实验结果表明EBP神经网络在双指标正交实验设计与分析中有较好的离线学习功能，可以较少的实验次数而获得更多的信息，指导实验设计与分析。

the synthesis of phosphorus free builder was studied using acrylic acid and maleic anhydride.The neural networks is trainedby the results of orthogonal design.The influences on the performance of build detergent by monomer ratio,initiator amount,reactiontime,monomer dripping time and reaction temperature are predicated by the neural networks.The best synthesis conditions are predi-cated by neural networks.The results show that EBP neural network has better off-time learning performance for two indexes of designand analysis of orthogonal experiments.Obtained more useful information by less experiments,and to guide the design and analysis oforthogonal Experiments.