A data-driven soft-sensing model of matte grade in oxygen-rich side blow bath smelting process

A data-driven soft sensing technique is proposed to predict the matte grade, which is aimed at the problem of delay in traditional methods for sampling and determining the matte grade in oxygen-rich side blow bath smelting (OSBS), resulting in the inability to control and optimize the process in real time. This method is based on real-time detected process data and utilizes particle swarm optimization (PSO) algorithm to optimize dynamic process features to construct the optimal input feature set. For the datasets after feature selection, a kernel based extreme learning machine (KELM) is used for regression model training. To adapt to the instant variations and dynamics of the copper smelting process, a local learning strategy combining sliding windows is introduced, which in turn builds an instant learning kernel extreme learning machine regressor. This instant learning regressor is particularly suitable for online learning and real-time prediction due to the simplicity, efficiency, and low time consumption of the KELM algorithm. The feasibility and effectiveness of the proposed method are verified through experiments conducted with data collected on an actual smelting production line.