Abstract: Based on the development of an online prediction model for the silicon content in blast furnace hot metal, this paper investigates the correlation between the lag time of operating parameters and silicon content in blast furnace hot metal, and proposes an online prediction model for silicon content in blast furnace hot metal based on long short-term memory(LSTM) networks. First, based on the production mechanism and the migration pattern of silicon, the lagging behavior of various parameters within the furnace is analyzed. Next, a database of silicon-influencing parameters with different lag times is constructed, allowing the determination of accurate lag times by analyzing the correlation strength between each parameter and silicon content across different lag times. Finally, based on data from 60 days of continuous production at a steel plant′s No. 4 blast furnace(a total of 16 131 samples), 10 operational parameters with high correlation are selected for predictive modeling through data preprocessing and feature selection. The LSTM-based online prediction model is designed to predict the silicon content in hot metal for the current time, 1 hour ahead, and 2 hours ahead. Results indicate that the online model demonstrates robust generalization ability, robustness, and accuracy, achieving a prediction accuracy rate of 94.6% within a 0.05% error range between predicted and actual silicon content values at the current time.