A deep learning model-based framework for abnormal state monitoring of blast furnace heat loads

Heat load is a pivotal metric for ensuring the safety of blast furnace equipment in the iron and steel smelting industry, crucial for monitoring the heat exchange process of blast furnace cooling walls and evaluating the erosion of furnace bodies. The variability of blast furnace heat load, influenced by factors such as feed and oxygen supply, often results in abnormal internal furnace temperatures and drastic heat load fluctuations, hindering effective preventive maintenance. This study introduces a deep-learning-based framework for monitoring the abnormal state of blast furnace heat load. Initially, a two-stage data preprocessing method for monitoring ironmaking parameters is proposed: the first stage eliminates missing and abnormal values from the monitoring parameters, while the second stage employs a feature selection model to identify key influencing parameters. Subsequently, a real-time prediction model of blast furnace heat load is designed, integrating one-dimensional convolutional neural networks (1DCNN) with bidirectional long short-term memory (BiLSTM) networks. This model uses a two-layer 1DCNN for automatic feature extraction from time series data, complemented by a single-layer BiLSTM network to accurately predict anomalous trends in heat load. This deep learning framework has demonstrated superior performance in prediction accuracy and abnormal trend identification compared to alternative models.