An effective control method for silicon content in hot metal: Model predictive control (MPC) based on CNN–LSTM–Attention algorithm

Abstract

As a key performance indicator in the smelting process of blast furnaces, silicon content in hot metal is used to characterise the thermal state and changing trend of the hearth. The silicon content in hot metal is closely related to the furnace condition and hot metal quality. Whether the silicon content in hot metal is too high or too low, it can cause accidents in the blast furnace. Predicting and controlling the silicon content in a hot metal can help avoid production accidents. Abnormal fluctuations in silicon content can be prevented. This ensures the safe and stable operation of the blast furnace smelting process. This study integrates the theory of the blast furnace smelting process and on-site data. It aims to predict and control the silicon content in hot metal by integrating process experience, big data technology and artificial intelligence. Firstly, data collected from the site was systematically integrated and processed. Subsequently, prediction models (XGBoost, LSTM, CNN–LSTM CNN–LSTM–Attention) were developed to predict the silicon content in hot metal for the next hour. Finally, with the help of MPC controller, an intelligent control model for the silicon content in hot metal was constructed to stabilise the silicon content in hot metal within the range of 0.25%–0.4%, providing a theoretical basis for achieving closed-loop control of the silicon content in hot metal.