Machine learning-based multi-objective parameter optimization for indium electrorefining

Abstract

A novel approach utilizing support vector regression algorithm (SVR) is presented for developing forecast models of Cu and Pb concentrations in indium electrolysis products. These models are based on a subset of process parameters and purity data. The optimization of Cu and Pb content is achieved through the integration of the forecast models with a multi-objective genetic algorithm. Consequently, a set of optimal electrolysis process parameters is identified for the electrolytic refining of high-purity indium. The determined optimal parameters are as follows: In3+ concentration of 80–90 g·L-1, NaCl concentration of 85–120 g·L-1, gelatin concentration of 0.5–0.6 g·L-1, current density of 65–70 A·m−2, pH value of 2.5, and pole pitch of 40–60 mm. To validate the effectiveness of these optimized parameters, experimental tests are conducted to confirm that the Cu and Pb contents conform to the national standard for 5 N indium. By employing this innovative approach, the study not only provides insights into the forecast modeling of Cu and Pb concentrations in indium electrolysis products but also contributes to the advancement of the electrolytic refining process for achieving high-purity indium.