基于广义回归网络的铝电解阳极效应预报

Research of Anode Effect Prediction of Aluminum Electrolysis Cell Based on Generalized Regression Neural Network

应用广义回归神经网络对当前预焙槽铝电解阳极效应预报问题进行了研究。在简述广义回归神经网络的基本结构基础上,利用广义回归神经网络对铝电解槽阳极效应进行系统辨识建模。重点探讨了建模过程中模型样本结构的选择,实验分析了样本容量对模型预报准确率的影响。取自某铝厂400 k A大型预焙槽的单槽运行现场数据样本对模型进行训练和检验,结果表明该方法阳极效应预报准确率平均在90%以上,预报提前量可以达到半个小时。现场多台电解槽的建模测试结果进一步论证了该模型和样本结构的合理性和有效性,由此证实该方法在保证较高预报准确率同时,具有较好的普适性。

This paper aims to research the anode effect （AE） prediction of pre-baked aluminum electrolysis cell with the generalized regression neural network （GRNN）. The structures and advantages of GRNN are introduced, then the anode effect system of aluminum electrolysis cell is modeled by the method of system identification based on GRNN. The structure of samples is analyzed emphatically in the process of modeling, and the influence of sample size to model prediction accuracy is analyzed by experiments. The AE model based on GRNN is trained and tested by sufficient samples which are extracted from the production data of the 400 kA aluminum electrolysis cell. It＇s proved that the accuracy rate of the AE prediction is more than 90% on average while predicting AE half an hour before the AE moment through this method. Results from experiments with sample data from different cells show that this analytic method is logical and effective, and has extensive applicability while keeping high prediction accuracy.