Abstract
The superheat of an electrolyte is an important indicator of the heat balance state of aluminum reduction cells. In industrial practice, it costs too much to accurately measure the superheat in every cell every day. A common alternative is to calculate the superheat based on additive concentrations in the electrolyte, which has problems of high error and long delay. In this paper, a method to diagnose the heat balance state of an aluminum reduction cell based on Bayesian network is presented, a Bayesian network structure and CPT (conditional probability distribution) were built, and the continuous diagnosis process is presented. This diagnosis method takes important symptoms and factors into account, taking advantage of more useful information instead of only calculated superheat. The application examples show that this method is effective in diagnosing the heat balance state for uncertain and incomplete superheat information.
Highlights
Maintaining a proper heat balance state is of great importance for the highly efficient and stable operation of aluminum reduction cells
It is quite suitable for diagnosing the heat balance state of aluminum reduction cell with uncertain input variables
A heat balance state diagnosis method based on Bayesian network is proposed
Summary
Maintaining a proper heat balance state is of great importance for the highly efficient and stable operation of aluminum reduction cells. Technicians in most aluminum smelters make indirect estimates of superheat by applying the following methods: sampling electrolyte and making the electrolyte composition analysis in the laboratory, calculating the liquidus temperature according to the empirical formula, and subtracting the liquidus temperature from the electrolyte temperature to obtain the superheat This method is not good in time-effectiveness, and has large errors in analyzing electrolyte composition. The development of the machine learning model in recent years has led many scholars to apply it to the prediction of the superheat, electrolyte temperature, and control system in aluminum reduction cells. Artificial neural network has received much attention from many researchers, there are many limits of application in predicting heat balance state due to the high technicality and complexity of artificial neural network. This paper analyzes the causality between the related variables of the heat balance state, on the basis of which the Bayesian network is established, calculating the conditional probability distribution from the manual measurement and online measurement data, proposing the specific continuous diagnosis procedures
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
