Prediction Of Aluminum Fluoride Addition And Aluminum Output In Aluminum Electrolysis Process Based On LSTM Neural Network

The addition of aluminum fluoride and the output of aluminum are two key indexes that affect the heat balance and material balance in the process of aluminum electrolysis.Therefore,it is of great value to study an accurate and effective decision method to obtain the amount of aluminum fluoride and the amount of aluminum produced.At present,although the automation level of electrolytic aluminum production has made great progress,the experience dependence of some key parameter settings on production managers is still high,and the production efficiency is easily affected by subjective factors.Therefore,using data mining technology according to historical data It is of great practical significance to predict the future as a research direction.A comparative study of the current decision-making methods on the amount of aluminum fluoride and the amount of aluminum produced is carried out in this paper.The current research status and shortcomings in this field are summarized,and the prediction research is carried out by using the LSTM neural network algorithm.Based on the understanding of the characteristics of aluminum electrolysis data and the coupling relationship of each parameter,the missing and abnormal values in the production data were pretreated,and then,in order to avoid overfitting and improve the efficiency of the algorithm,the characteristics of the processed data were selected by random forest algorithm,and the first 8 features were obtained,which were highly correlated with the addition of aluminum fluoride and the output of aluminum,and were used as LSTM neural network The data set is normalized and divided into three data sets: training,prediction and verification.On the training and test set,the optimized network structure and hyperparameters are adjusted continuously,the appropriate loss function and optimization function are selected.Finally,the final prediction is carried out in the verification set,and the results can meet the actual production requirements are obtained,which prove the application value of the algorithm.Lastly,using the Python language and the third party tools Py Qt5,a software of aluminum electrolysis parameter prediction system which can realize the visualization of the whole process of the above algorithm and operate conveniently is designed and developed in the Eric6 development environment.The software includes: login registration,data loading,data analysis,data processing and neural network and other functional units.