Tumble Strength Prediction for Sintering: Data-driven Modeling and Scheme Design

Tumble strength (TS) of sintered ore essentially determines the smooth operation of the blast furnace and the quality of the iron. However, due to the lack of timely measurement, TS is hard to estimate and predict, resulting in difficulty for accurate control. Thus, it is a challenging problem to realize precise prediction models. The physical and chemical reactions in the sintering bed are hard to describe, incurring high complexity to precisely model the dynamics. In addition, due to the limited features contained in the data acquired from the sintering site, the general scheme of the data-driven method cannot regress well. Moreover, the intrinsic delay in the sintering and measuring process causes inevitable time-tag mismatching of input and output data, further degenerating the prediction model performance. To tackle these issues, in this paper, we first propose a data-driven TS prediction scheme combined with a local thermal non-equilibrium (LTNE) model, which deals with the uncertainty in the thermo-chemical reaction equations of the sintering bed. Through the sinter pot tests, test data is then applied for feature extracting for accurate modeling. Lastly, The data time-tag matching is executed by inferring the sintering process and measurement delays. The accuracy of TS prediction reaches 99.49% compared with the real field data, which demonstrates the effectiveness of the proposed TS prediction scheme.