Predictive Analytics for Enhancing Productivity of Reduction Cells

Abstract

With increasing energy prices and lower LME, aluminum smelters across the globe are focusing on reducing specific energy consumption. Increasing current efficiency (CE) not only reduces energy but also increases productivity. Since, early 1990s, fundamental studies and lab-scale experiments have provided insights on CE and its dependence on process parameters, however these were based on ideal conditions and actual plant data should also be considered. This article presents a predictive model for CE utilizing machine learning algorithm (random forest regressor) on 360 kA pot-line data. The model helps in identifying the optimal parameter range to maximize CE of individual pot. Results are compared with fundamental and lab-scale experiments published in literature, showing good agreement in most cases along with few insights. Impact of parameters such as cathode drop, bath height, composition, etc. has been discussed.