Optimization Of Aluminum Electrolysis Process Parameter Range Based On Deep Reinforcement Learning

In the electrolytic aluminum production process,process personnel usually need to adjust the optimal control range of a series of process parameters such as aluminum level and molecular ratio based on experience to ensure the stable and efficient operation of the electrolytic cell.At the same time,the combination of different ranges of many parameters may reach the same Control effect.In response to this problem,this article explores the use of deep reinforcement learning algorithms on the basis of a large amount of historical production data of electrolytic aluminum to find and determine the optimal value range combination of each process parameter of the electrolytic cell.The main research work of the thesis is as follows:1)The original data generated during the production process of the electrolytic cell of an aluminum factory is preprocessed with missing value completion,abnormal value screening,and normalization,and the original data and processed data are visually displayed.2)Use the two sigma method to optimize the parameter range of a single parameter.3)The average aluminum output is used to mark the daily historical production data of the electrolytic cell,and the C4.5 decision tree algorithm is used to obtain the range combination of process parameters.4)The deep reinforcement learning algorithm model based on Proximal Policy Optimization(PPO)is used to optimize the range of electrolytic aluminum parameters.By adding external rewards and punishments,the convergence speed of the model is accelerated,and the root mean square error(RMSProp)is introduced into the near-end strategy optimization(PPO),and then the momentum is used to correct the cumulative gradient of each iteration.The improved algorithm M-RMSProp-PPO has faster convergence and higher reward value.5)Perform clustering based on spatial noise density(DBSCAN)on multiple sets of optimization results of M-RMSProp-PPO,and merge the results of decision tree division to determine the final optimal parameter range combination.6)Designed and implemented an aluminum electrolysis process parameter range optimization system based on deep reinforcement learning.