A Novel Superheat Identification of Aluminum Electrolysis with Kernel Semi-supervised Extreme Learning Machine

Abstract

In the aluminium reduction production industry, the superheat temperature (ST) is a vital index, which gives the distribution of the current efficiency. Keeping ST in an approximate range improves the lifespan of the electrolysis bath. However, in the practical production process, ST identification result is commonly evaluated by the experimental workers, the real-Time measurement of ST is still a challenge that has not been solved. A novel ST measurement method based on kernel extreme learning machine (K-ELM) is studied in this paper. First, a few of input variables are selected according to expert experience. Then, a new activation function (IG kernel function) and regularization term are constructed in ELM to construct IG-SSELM. Finally, the ST model is built with all the dataset samples and further applied to ST real-Time detection. The proposed method is applied to ST identification for the first time in the aluminium industrial process which is superior to the existing ST methods in accuracy and robustness.