Distributed Anode Current Classification Method Based on Spectral Filtering and Convolutional Neural Networks

Distributed anode current (DAC) is a spatio-temporal sequence containing important spatial distribution information that can be obtained online by the aluminum electrolysis cell control system. DAC classification is of great significance in the knowledge discovery of aluminum electrolysis industrial production process data. Due that DAC has the dynamic cell noise and complex temporal-spatial dependency characteristics, DAC classification is a challenging problem and existing classification methods cannot effectively capture these spatial features under the interference of industrial noise. To solve the above problems, DAC classification method based on spectral filtering and convolutional neural networks (CNN) is proposed in this paper. Firstly, the dynamic graph structure of aluminum electrolysis cell is obtained based on the similarity between each dimensional signal of DAC to capture the dynamic spatial dependence of DAC. Secondly, taking into account the real anode spatial structure information (ASSI) and the feed mode of the aluminum electrolysis cell, a static graph structure is designed to capture the static spatial dependence of the DAC. Then, two different spectral filtering modes are proposed, and the spectral filtering results of the DAC are converted into gray-scale images, which are used as the input of the CNN to achieve the DAC classification task. Finally, the classification results on the actual DAC datasets compared with existing methods demonstrate the effectiveness of our method.