Classification of Anode Current Signals Based on 1D Convolutional Neural Networks

Intelligent and refined production becomes the development direction of the aluminium electrolysis industry. Anode current signals (ACS) can reflect the local conditions of electrolytic cells, timely and accurately classify the anode current signals, which will help to achieve regionalization and fine control of cells. Anode current signals are typical multivariable time series, so it is difficult to obtain its discriminant features based on traditional spectrum classification methods. Therefore, this paper presents a method to classify the anode current signals using one-dimensional convolutional neural networks (1D-CNN). In addition to the input layer and output layer, the proposed CNN model consists of 8 layers, including 3 convolution layers, 2 max-pooling layers, and 3 fully connected layers. The model can automatically extract the features from the original data, so as to realize the three types of anode current signals classification, namely, normal, anode effect (AE) and anode change (AC). The experimental results show that the classification accuracy reaches 87.6%, which verifies the effectiveness of the method.