A Dynamic Control Model of Basic Oxygen Furnace Last Blowing Stage Based on Improved Conditional Generative Adversarial Network

In the basic oxygen furnace (BOF) steelmaking process, the last blowing stage occurs between the first and second sub-lance measurements and utilizes the dynamic control model. Traditional dynamic control models rely solely on the detection value by TSC (Temperature sampling carbon) to set process parameters for the last blowing stage. This approach neglects the strong coupling relationship between time series data, such as lance height and bottom gas supply intensity, leading to inaccurate control of endpoint carbon content and temperature. To address the above problem, a dynamic control model of BOF last blowing stage based on improved conditional generative adversarial network (ICGAN) is proposed. The ICGAN model incorporates data prior to TSC as additional conditions for collaborative analysis, including properties of raw materials and time series data like process parameters from the main blowing stage. The time series data mainly include lance height, oxygen supply intensity, bottom gas supply intensity, and coolant addition amount. Furthermore, the model includes a supervisor, in addition to the generator and discriminator, to establish the relationship between process parameters and endpoint property indicators. Therefore, the ICGAN model can be used to conditionally and in a supervised manner generate process parameters for the last blowing stage to obtain molten steel that meets the property requirements. The effectiveness of the ICGAN model is evaluated using actual data from two steel grades produced in BOF steelmaking. The hit rates for endpoint carbon content of the two steel grades within an error range of ±0.015 wt pct are 91.64 and 91.81 pct, respectively. The hit rates for endpoint temperature within ±15 °C are 88.57 and 89.98 pct, respectively. Additionally, the double hit rates for both carbon content and temperature are 87.35 and 88.15 pct, respectively.