Prediction model of the end-point carbon content in BOF by constructing the graph of heats

抄録

For the differences in composition and proportions of input materials, end-point quality requirements and slagging specifications, it is difficult to construct a generalized model to guide steelmaking production. There is a complex connection among different heats in BOF, and this combination of correlations can be thought of as a hidden working pattern. A method named "attribute level division" is developed to exploit the correlations among heats to construct the graph of heats. The model based on the combination of label propagation algorithm (LPA) and back propagation neural network (BPNN), LPA-BP, is proposed for end-point carbon content prediction in BOF. LPA is used to discover the different community in the graph of heats and BPNN is trained to construct different models for end-point carbon content prediction for the heats from different community. The results of comparative experiment show that the LPA-BP model is higher than the baseline 2.5% when prediction error is within ±0.012%. The LPA-BP model also outperforms in some metrics, such as RMSE, MAE. This model provides a novel idea to improve the endpoint hit rate by distinguishing different communities to uncover the hidden working patterns among heats and constructing different models.