Research On Production Decision Algorithm Of Aluminum Reduction Cell Based On Deep Learning

Optimizing the aluminum electrolytic production decision can produce huge economic benefits,but due to the characteristics of non-linear,large lag,multivariable coupling and other characteristics of the aluminum electrolytic production process,it is difficult to make the optimal decision through manual analysis by the process personnel alone.During the years of aluminum electrolytic cell production control and management,major aluminum electrolytic plants have collected a large amount of production control,measurement,testing,and decision-making data.Using these historical data to train machine learning systems is an important way to solve aluminum electrolytic decision-making optimization.Deep learning is a machine learning method that has recently made breakthrough progress.It is characterized by learning high-quality features from a large number of samples and is widely used in machine vision,natural language processing,robot control and other fields.In this paper,a deep learning model based on the internal node bagging algorithm is used to predict the optimal aluminum production decision of the electrolytic cell under different states,so as to maximize the cumulative output of the electrolytic cell.The main research contents of the paper are as follows:1.Preprocessed and visualized the raw data of electrolytic production of an aluminum plant.First,the original data is analyzed for missing and abnormal conditions,and then the data is cleaned through sequence extraction,outlier processing,and missing value processing,and the data cleaning effect is demonstrated through various charts.Finally,the distribution of each feature of the data and the correlation between them are analyzed,and each feature is scored using xgboost.2.The Markov Decision Process(MDP)was used to mathematically describe the aluminum electrolytic cell production process,which made the research work separate from the complicated aluminum electrolytic process.An electrolytic cell simulation algorithm based on KNN(K nearest neighbor,k-Nearest Neighbor)is proposed,which lays the foundation for the subsequent deep model training and evaluation in this paper.3.A deep decision model based on the inner node bagging algorithm was proposed,and it was trained using supervised learning and reinforcement learning.To this end,an aluminum electrolytic cell condition marking algorithm and a simulator-based DDPG reinforcement learning algorithm were proposed,and The performance of the benchmark model,the supervised model and the enhanced model are compared and analyzed on the simulator.4.A decision system for aluminum electrolysis was designed and implemented.