热分解过程的神经网络预测方法的研究

提出了用前馈神经网络求解热分解过程的新方法．针对传统的ＢＰ算法的缺陷，把无约束优化中的变尺度应用于网络的训练学习，改进了学习算法，提出了一种基于动态步长的新的变尺度算法．通过它对热分解过程的预测分析，其结果非常逼近实验结果并优于传统的理论计算结果；同时，新的变尺度算法提高了网络算法的收敛性．

Study on the Neural Networks Method to Predict Thermal Decomposition

A new method for the neural networks to calculate thermal decomposition is presented. In view of the defects of the traditional BP algorithm, variable metric in unconstrained optimization is applied to train the neural networks, thereby the learning algorithm is improved. At the same time a new variable metric algorithm based on the dynamic stepsize is proposed. The algorithm is used to predict thermal decomposition. The results show very good agreements with the experimental data, which indicates that the neural networks method can be used to predict thermal decomposition with better accuracy than the traditional theory. The new variable metric algorithm improves the convergence of the neural networks.