Research And Application Of Alumina Dissolution Temperature Decision-making Based On Deep Reinforcement Learning

The high-temperature dissolution process of alumina is an important part of alumina production,which is characterized by the interaction of many factors.Among them,the dissolution temperature is the most important process index that affects the quality of the dissolution.It is difficult for the production personnel to achieve accurate decision-making and control of the dissolution temperature..Through the research on the actual process flow of the dissolution link of alumina enterprises and the analysis of a large amount of historical data,this paper adopts the deep reinforcement learning algorithm to make decisions on the dissolution temperature of the dissolution link of alumina,so as to improve the dissolution quality.(1)Missing value,outlier processing and visual analysis were carried out on the original production data related to the dissolution link provided by the enterprise,and the importance of each parameter characteristic in the alumina dissolution link was analyzed by using the XGBoost algorithm.(2)Through the study of the target network parameter update strategy,a deep reinforcement learning algorithm based on the dynamic update of the target network is proposed.Aiming at the shortcomings of the experience replay mechanism,a deep reinforcement learning algorithm based on K-means classification experience replay is proposed.In the classical test environment,compared with the traditional deep reinforcement learning algorithm,the convergence speed and post-convergence stability of the improved deep reinforcement learning model are greatly improved.(3)The mathematical model of the alumina dissolution process was established using the Markov decision process.According to the actual production historical data,a neural network-based simulation algorithm for the alumina dissolution process was designed to provide support for the environmental state transition of the subsequent decision-making model.(4)An alumina dissolution temperature decision model based on improved deep reinforcement learning is established.The experiments show that the alumina dissolution temperature decision model based on improved deep reinforcement learning is better than the traditional linear model and the traditional deep reinforcement learning model.(5)A decision-making system for alumina dissolution temperature based on deep reinforcement learning was designed and developed.The system provides functions such as data visualization analysis,decision model training,decision model application,and decision model evaluation.