Anode Effect prediction based on Expectation Maximization and XGBoost model

Abstract

Anode Effect Prediction problem has been drawing great research interest of scientists, due to its significant values in reducing energy consumption and improving the efficiency of aluminum electrolysis. However, a large number of missing values contained in the collected data from the aluminum reduction cell are always neglected in the works, resulting in a decline in prediction accuracy and generalization ability. To solve this problem, a combined model of Expectation Maximization and XGBoost (EM-XGBoost) is proposed. Firstly, the original incomplete samples collected from the aluminum cells are recovered by Expectation Maximization (EM) algorithm. Afterwards, the XGBoost model trains on the recovered data, and then predicts the result for new samples. The more comprehensive metrics accuracy and F1 Score are introduced for evaluation. The results in the experiment show that the proposed model improves the accuracy to 99.7% and the F1 Score can achieve 99.8% under the premise of forecasting 30 minutes in advance. The proposed model not only has a high prediction accuracy, but also owns an excellent generalization ability.