神经元网络理论在乙烯精馏塔成分估计中的应用

针对前馈神经元网络误差反向传播算法（ＢＰ算法）收敛速度慢，且常常收敛于局部极小值等缺陷，提出了一种基于变步长、加压缩因子的共轭梯度搜索的快速学习算法。与标准的ＢＰ法相比，该方法不仅学习收敛速度快，而且精度也有所提高。通过对乙烯精馏塔成分估计的应用表明，该方法比最小二乘估计法具有更好的外推性。

Application of Neural Network Theory in the Estimation of Ethylene Distillation Column Components

Based on the search of conjugate gradient which has changed steps and com-pressibility factors,a new fast learning method is proposed countering the drawbacks of theslow convergent speed and the local minimun of BP method. By comparison with BP methodalgorithm, the new method has not only faster learning speed but also higher accuracy,Itshows that the exptrapolation of the new algorithm is better than that of the least squaremethod in estimation some distillation column.