A Deep Learning Method for the Endpoint Carbon Prediction in BOF Steelmaking Process

In many basic oxygen furnace (BOF) steelmaking processes, if the furnace endpoint carbon can be monitored in real time, it is a breakthrough for BOF steelmaking intelligence. This paper presents a deep learning model used to predict the endpoint carbon content in BOF steelmaking process. A convolution long short-term memory network based on attention mechanism (CNN-LSTM-AM) model is proposed for time-series data in BOF process to extract spatial-temporal characteristics of time sequence features and a back propagation (BP) model is proposed for pre-furnace data to auxiliary increase the accuracy of the model. The BOF steelmaking data from an actual process were used for the testing, the result shown that 84.34% of prediction result were within the ±0.02 range, which is better than use those two types of data and model individually.

1 Introduction

BOF steelmaking process is using iron, slag, and other materials, relying on the heat carried by the raw materials in the converter as well as the chemical reaction exothermic heating to remove carbon, phosphorus, and other elements, increase the iron content and ensure that the appropriate temperature of the steel process. As an important part of steelmaking, the stability and accuracy of the converter endpoint control is crucial to the steel quality control in the entire process. If the blowing time is too short, it will cause high carbon content at the endpoint, will result TFe (total ferrum) increase in the slag, increase metal consumption, reduce the life of the furnace wall, and need further blowing to correct. If the blowing time is too long, it will cause low carbon content at the endpoint, and need further refill the carbon. Both high and low carbon content at the endpoint affects steel production quality and increases the waste of raw materials and energy.