作　　者： ; ; ; ; ; ; (田德余）;

机构地区： 深圳大学师范学院化学生物系

出　　处： 《火炸药学报》 2004年第2期1-6,共6页

摘　　要： 用误差反向传播(BP)的人工神经网络模型及分子结构描述码作为输入特征参数预测非芳香族多硝基化合物的生成焓,研究了网络参数及分子结构描述码的影响,同时按分子结构描述码进行了多元线性回归,取得了满意的结果,其回归方程相关系数达到了0.9977,精度高于文献值。绝大多数相对误差在10%以内。 With the help of an improved back propagation algorithm on Artificial Neural Network (ANN) model and molecular structure descriptors as inputted characteristic parameters were used to predict the enthalpy of formation ΔH^0f for non-aromatic polynitro compounds (NAPNCs). The influence of Neural Network parameters and molecular structure descriptors on the calculated results were studied. Satisfactory result was made in the multiple linear regression (MLR) calculation of molecular structure descriptors. The correlation coefficient of MLR equation for 70 samples is 0.9977. The accuracy of the enthalpy of for mation calculated by ANN is higher than reported value in reference. The absolute relative difference between reported and predicted values of ΔH^0f is within 0.10 for most of NAPNCs.

关 键 词： 人工神经网络 生成焓 分子结构描述码 非芳香族多硝基化合物

领　　域： [理学] [理学]