A Cell Voltage Classification Method Based on Wavelet Packet Analysis and Residual Network

摘要

In the aluminum reduction production process, the cell voltage signal is an influential index of the cell condition, reflecting the stability and controllability of the aluminum reduction cell. According to the spectral characteristics of the voltage signal of aluminum reduction cell, in this paper, we propose an integrated method for classifying and evaluating the cell voltage signal. Firstly, we decompose the original signal into low frequency components and high frequency components using wavelet packet analysis and we construct a energy model of the cell voltage signal for the calculation of the proportion of high frequency fault energy consumption of the original voltage signal. Then we build a dataset of low frequency component samples of cell voltage signal and we train a residual network model on this dataset for the identification of the fluctuations of low-frequency component of cell voltage. Finally, some expert rules are developed to determine the state of cell voltage. And the experimental results show that the effectiveness of this classification method for the analysis of the cell state.