Research On Molecular Ratio Trend Prediction Of Electrolytic Tank Based On Deep Learning

In the process of aluminum electrolysis production,intermediate parameters such as molecular ratio parameter play an important role in responding to the quality of aluminum solution,material balance and current efficiency of the electrolyzer in the current state.By predicting these parameters and their trends,we can better achieve accurate control of the electrolytic cell,further reduce energy consumption,improve current efficiency,and promote green and intelligent aluminum electrolysis production.This paper firstly introduces the main processes and technical parameters of the electrolytic aluminum production according to the production principle of modern electrolytic aluminum process,and then starts the targeted data pre-processing work according to the main characteristics of the actual production data used in this project.Secondly,we propose a combined Attention-CNN-LSTM model for molecular ratio prediction of electrolytic cells based on long and short-term memory networks by introducing the attention mechanism and convolutional neural networks to address the characteristics that the input variables are coupled with each other in the molecular ratio prediction work and the fluoride salt addition has a large impact on the final results.The model uses the attention mechanism and CNN module to extract features from the original input data,and the results are used as the input to the LSTM,and the prediction model is built to generate the prediction results based on this.Again,for the characteristics that the electrolytic aluminum production process data has a certain lag and the LSTM gating unit structure is large,this paper introduces CNN,attention mechanism and bi-directional gating unit,and proposes an Attention-CNN-Bi GRU molecular ratio prediction model.The model uses CNN algorithm to extract the local features of the input data,and feeds its extraction results into the Bi GRU layer with the attention mechanism added to achieve the molecular ratio prediction work.For the molecular ratio trend prediction,the molecular ratio parameter values and predicted values over a continuous period of time are selected in this paper,and the least squares method is used to perform a grouped linear fit for the actual and predicted molecular ratio parameter values over a fixed time interval by minimizing the sum of squares of the residuals,using multiple groups of different time intervals.The trends of the molecular ratio parameter values over the time interval are described by the linear equations corresponding to the actual and predicted values in different groups and their derivatives.In this paper,experiments were conducted using actual production data,and the GRU,Bi GRU,LSTM,Attention-CNN-Bi GRU,and Attention-CNN-LSTM prediction models were evaluated using root mean square error and mean absolute percentage error.For the molecular ratio trend prediction,the actual and predicted values of the molecular ratio parameters within 21 consecutive days were selected in this paper,where the actual values were 14 days and the predicted values were 7 days,and the prediction models were grouped at 3-day and 7-day intervals,respectively,to evaluate the accuracy of the prediction models in trend prediction.The experimental results showed that the above prediction models were able to predict the molecular ratio and trend of the electrolyzer more accurately,thus providing guidance for accurate control of the electrolyzer.Finally,by using the improved algorithm,a molecular ratio prediction system for the electrolyzer was designed and developed in this paper.The system can provide accurate mole ratio prediction and support the decision making of electrolytic aluminum production by implementing the functions of data reading,data analysis,data pre-processing,prediction model training,model evaluation and prediction.