ON THE USE OF LADLE-WISE ANALYSES TO PREDICT HOT METAL SILICON CONTENT

摘要

A method to quantify and track the tap-internal trends that can be observed in hot metal silicon content has been introduced. A simple polynomial used to quantify the trend was found to adequately describe the characteristics of the general, varying, trends. The method can also be useful as a means to monitor changing conditions that can affect the internal state of the blast furnace hearth. Furthermore, the method was employed as a filter to the ladle-wise analyses of silicon content and this was seen to be beneficial for the development of all empirical predictive models presented. For the (time varying) linear ARMA models the benefit of the filter was more apparent and a possible explanation could be that the filter addresses fundamental nonlinear aspects of the signal. In order to explore these findings, the residuals for all models were analysed, albeit with quite inconclusive but also very intriguing results, thus motivating future efforts to shed light on this topic. For the practical use of the models presented, it could be desirable to combine the stable characteristics of the (possibly nonlinear) time series models with the potential for predicting sudden changes inherent in FIR models. A simple combination of the inputs in a type of model commonly known as NARMAX (nonlinear auto regressive moving average with exogenous inputs) model is not, however, an attractive approach because of the possible inertia due to the NARMA-part introduced in the FIR structure. The greater flexibility of using multiple models, e.g., an NARMA and a possibly nonlinear FIR model respectively, with an "intelligent" criterion for model selection, seems a more promising possibility and this option will be explored in future work. An interesting possibility is also that, given a better description of flow conditions in the hearth, alternative orderings of the ladle-wise analyses could be considered. For this purpose, future research aims at developing a (simplified) model for flow patterns in the hearth. In addition to possibly providing a physically motivated ordering of the ladle-wise analyses, such a model could also supply more realistic time-labels for the observations and the latter could improve on the quality of the FIR models.