Modelos de séries temporais para previsão de nível de líquidos em cadinho de altos-fornos

Abstract

The operation of material extraction from blast furnace is carried out with a significantdegree of uncertainty, among other reasons, because the measuring level of liquids cannotbe measured directly. This thesis presents a system for forecasting the level of liquid in the blast furnace hearth by measuring the electromotive force generated in shell based on a model seasonal autoregressive integrated moving average (SARIMA). This work has shown electromotive force is a non-stationary and nonlinear time series with a strong seasonal behavior that is strongly correlated with the level of liquids. Some comparisons were made with models based on artificial neural networks with time delay (TDNN) and the results indicated that the nonlinear model has better forecasting performance. This methodology consists of the strategy for analysis, identification, filtering and prediction of the level of liquids through TDNN models achieving at the end of the process a prediction with satisfactory accuracy. The forecast level of liquids with horizon up to 1 hour ahead can help operators and engineers during the control and process optimization of the production of blast furnaces increasing safety and financial gains.