Research On Setting Voltage Of Electrolyzer Based On Deep Learning

In recent years,the production of electrolytic aluminum in China has been increasing steadily.In the process of aluminum electrolysis production,the setting voltage of electrolytic cell plays an important role in the stability of production.Real-time monitoring and optimization of cell voltage changes can further reduce energy consumption and improve current efficiency,which has a profound impact on electrolytic aluminum production.According to the principle of electrolytic aluminum production process,this paper analyzes and preprocesses the historical data of electrolytic aluminum production from multiple angles.Secondly,GBDT algorithm was used to build a model for data,and the importance of its features was analyzed.Aiming at the problems of GBDT algorithm overfitting and slow learning speed,Light GBM algorithm based on leaf growth mode and histogram algorithm was proposed.In view of the characteristics of the electrolytic aluminum production process data,such as certain lag and multi-variable coupling,bi-directional LSTM was introduced.At the same time,the forward and backward hidden features were output as the output layer,and the attention mechanism was introduced to consider the influence of key features on the model results.Finally,Bi LSTM-Light GBM combined model algorithm is designed and implemented.Several algorithms were evaluated by using the root mean square error,mean absolute percentage error and determination coefficient R2.The accuracy of the combined model algorithm is better than other algorithms,which proves the effectiveness of the proposed algorithm.Finally,the voltage setting decision system of electrolytic cell is designed and developed by using the improved algorithm.The modules of data reading,data visualization analysis,data preprocessing,model training,model application and model evaluation are realized.Provide decision-making suggestions for process technicians and apply them in actual project development.