采用BP神经网络研究C-F键核自旋偶合常数

通常理论研究核自旋偶合常数的方法是基于线性模型进行拟合和预测，该方法在拟合和预测中仍有较大误差. 本文在前面工作的基础上，提出了基于非线性模型对C-F键核自旋偶合常数进行研究的观点，采用BP神经网络方法对C-F键核自旋偶合常数的函数关系式进行拟合，并用拟合结果对4种化合物的偶合常数进行预测. 结果表明，采用非线性的BP神经网络方法其训练效果与预测效果均优于线性模型方法；其预测误差对文中的4种化合物不超过0.40%.

Calculation of Nuclear Spin-Spin Coupling Constants of C-F Bonds by A Nonlinear Model and Back Propagation(BP) Neural Network Analysis

Linear models are often used to describe the relationship between nuclear spin-spin coupling constant and structural parameters, despite it has been shown that calculation using these models often result in large errors. Based on the results of previous works, a non-linear model was proposed in this study to describe the relationship between the spin-spin coupling constants of the C-F bonds and the chemical environment they are in. Back propagation (BP) neural network analysis was used to fit the experimental data to the model. The accuracy of the model proposed was tested in four compounds, and it was shown that the non-linear model fitted by BP neural network analysis provides much better predictions than the commonly used linear models. Calculation errors for the four test compounds were less than 0.4% when the nonlinear model was used.