Hybrid event-, mechanism- and data-driven prediction of blast furnace gas generation

Abstract

Blast furnace gas (BFG) is a byproduct gas and a significant energy source in integrated steelworks. Precise BFG generation prediction plays a pivotal role in site energy scheduling and management. However, it is difficult to accurately predict fluctuations in BFG generation due to the variable operational statuses and complex chemical reactions that occur inside the blast furnace, hindering efficient energy utilization and accordingly causing BFG to flare and contribute to environmental pollution. To tackle this problem, a hybrid event-, mechanism- and data-driven prediction method is proposed in this work. In this novel approach, blast furnace operational events are considered when predicting BFG generation, thus making predictions more accurate by integrating a priori mechanism knowledge associated with the blast furnace ironmaking process; additionally, this approach ensures high accuracy by selecting the best available data-driven prediction model for different event-associated periods. To demonstrate the predictive performance of the proposed hybrid method, comparative experiments are conducted using practical data from integrated steelworks. The results highlight the excellent performance and accuracy of the proposed method when compared with the results of widely used moving average and artificial neural network models.