论文部分内容阅读
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。
The general regression neural network is applied to predict the anodic effect of aluminum electrolysis in prebaked cells. Based on the basic structure of generalized regression neural network, the generalized regression neural network is used to systematically identify and model the anode effect of the aluminum reduction cell. This paper focuses on the selection of the model sample structure in the modeling process and analyzes the influence of the sample size on the accuracy of the model prediction. The single tank operating field data from a 400 kA large prebake cell in an aluminum plant was used to train and test the model. The results show that the anomaly forecast accuracy of this method is above 90% on average, and the forecasting advance can reach half an hour. The results of modeling tests on multiple electrolyzers in the field further demonstrate the rationality and validity of the model and the sample structure. It is proved that the proposed method has good universality while ensuring the accuracy of forecasting.