A Dynamic Spatio-temporal Synchronous Spectral Filtering Method for Distributed Anode Current Signals Classification

In the modern industrial aluminum electrolysis production process, distributed anode current signals (DACS) classification is an effective means to mine knowledge from data. Considering that DACS is a spatio-temporal sequence signal with complex dynamic spatio-temporal dependencies that are difficult to be captured by traditional methods. To solve this problem, a dynamic spatio-temporal synchronous spectral filtering method for distributed anode current signals classification is proposed. Firstly, the edit distance on real sequences (EDR) distance function with strong robustness is used to calculate the similarity between one-dimensional anode current time sequence corresponding to different carbon anode positions and the similarity between one-dimensional anode current spatial sequence corresponding to different time points, and dynamic spatial graph and dynamic temporal graph are constructed, respectively. Secondly, dynamic spatial spectral filtering and dynamic temporal spectral filtering are performed synchronously on DACS based on dynamic spatial graph and dynamic temporal graph respectively to capture complex dynamic spatio-temporal dependencies. Then, based on the mechanism knowledge of industrial aluminum electrolysis production process, the attenuation intervention is performed on spatial spectral filtering to make the model adapt to different cell conditions, and the corresponding matrix is spliced with the results of dynamic temporal spectral filtering to obtain the feature input matrix for DACS classification. Finally, a pre-defined static graph is designed based on the electrolyte flow field distribution. Meanwhile, to prevent over-smooth issues, a single-layer graph convolutional networks (GCN) is used to perform DACS classification tasks. And, the classification results on the actual DACS datasets compared with existing methods demonstrate the effectiveness of proposed method.