A Method for Anode Effect Prediction in Aluminum Electrolysis Cells Based on Multi-scale Time Series Modeling

The aluminum industry is moving toward intelligent and low-carbon development. Accurately predicting the anode effect has always been a significant challenge in monitoring the aluminum electrolysis process. However, due to the high-temperature and high-magnetic detection environment of aluminum electrolysis cell, some critical parameters cannot be measured online. This inconsistency in data flow makes it challenging to apply traditional data-driven methods directly. In response to the characteristics of large data samples collected in actual production, we have proposed a multi-scale time series modeling approach based on hybrid deep learning. This method combines three advanced neural network models: BiLSTM, LSTM, and DNN. It enables the extraction of parameters that influence the anode effects from both short-term and long-term cyclic variables. Compared to traditional shallow machine learning methods, deep learning methods, and hybrid learning methods, our proposed algorithm achieves the highest accuracy and F1 score, reaching 0.95 and 0.93, respectively. These results hold significant promise for reducing energy consumption and carbon emissions in actual production processes, paving the way for future applications.