编码输入法在PCB分配参数预测中的应用

APPLICATION OF CODING INPUT STRATEGY IN PREDICTING PARTITION PARAMETERS OF PCBs

根据Free Wilson法中化合物结构表达的思想 ,采用两种简单的编码输入方法对 5 8个多氯联苯 (PCB)的结构进行表征 ,并基于模型简单性原则对多元线性回归 (MLR)与误差反向传递 (BP)人工神经网络、模拟退火 (SA)人工神经网络和遗传算法 (GA)人工神经网络PCB分配参数预测模型的预测能力进行了比较 ,试验证实 ,粗略考虑PCB结构对称性的简单编码输入规则可以简化PCB分配参数预测模型的数字形式 。

On the basis of structure characterization concept defined in Free-Wilson method,two simple coding input strategies were adopted in the structure description of 58 PCBs and different models were established to predict n-octanol/water paritition coefficients of PCBs by using various mathematic tools such as multiple linear regression,back propagation neural network,simulated annealing neural network,and genetic algorithm neural network,respectively.Furthermore,Comparison of prediction capactiy of obtained models was conducted.As a result,the simple coding rule considering the structural symmetry of PCBs showed amazing potential in simplifying the prediction models,and multiple linear regression model based on it presented better mechanism interpretation ability than the corresponding neural networks.