基于深度学习的氟化铝添加量和出铝量预测

Prediction of aluminium fluoride addition and output based on deep learning

本文利用数据挖掘技术,针对铝电解生产过程提出了一种对氟化铝日添加量和日出铝量进行预测的算法。通过搭建LSTM神经网络,将经由随机森林算法选取的强特征子集当日数据作为网络的输入,氟化铝添加量和出铝量第二日数据作为输出进行训练和测试,最后对连续10天的数据进行预测验证,得到氟化铝日添加量和日出铝量的平均绝对误差分别为1.32和25.21,能够满足工业生产要求。

Using data mining technology,this paper puts forward an algorithm to predict the daily amount of aluminum fluoride and the daily amount of aluminum.By building a LSTM neural network,the real-day data of strong feature subsets selected by random forest algorithm are used as the input of the network,and the second day data of aluminum fluoride addition and output are trained and tested.Finally,the average absolute errors of daily addition and sunrise aluminum are 1.32 and 25.21,which can meet the requirements of industrial production.