有序神经网络及在阳极效应预报中的应用

Ordered Neural Network and its Application to Prediction of Anode Effect

提出了利用有序神经网络研究铝电解槽阳极效应的预报问题。概述了铝电解槽及其阳极效应的基本情况,针对铝电解槽控制难题和传统方法的不足,选择有序神经网络用于阳极效应概率预报。描述了有序神经网络的基本结构、与传统单隐层BP神经网络的区别以及由此带来的网络映射性能的改善,并使用梯度下降原则推导了有序神经网络的学习算法。使用铝电解槽的现场数据对有序神经网络进行训练并检验,结果表明有序神经网络可以比传统神经网络更及时、准确地对铝电解槽阳极效应进行预报。

Ordered neural network （ONN） is applied to prediction of anode effect （AE） in aluminium electrolysis cell. The neural network （NN） and other methods in aluminum electrolysis cell fault diagnosis are reviewed, and the comparative advantages of NN method are analyzed. ONN topology structure is introduced and learning algorithm is derived. Improvements from traditional backpropagation NN （BPNN） to ONN are illuminated. Eventually, correctness and practicality of the application is validated. ONN and some typical NNs are trained and tested with using real data from aluminum electrolysis plant. In contrast with other NNs, ONN can predict aluminum electrolysis cell AE more timely and rishtly.