Artificial neural network-based committee machine for predicting fuel rate and sulfur contents of a coke blast furnace.

Abstract

Being developed over the centuries, it currently occupies a prominent role in the world production scenario, being the stage of the process related to the obtaining of hot metal an element of great importance to establish the competitiveness of national steel. From this perspective, the control of the process of obtaining hot metal is relevant to ensure competitive prices and a sustainable process. Considering the presented situation, this research developed a committee machine, being three networks to predict each of the study variables, namely: i) fuel rate; ii) sulfur content in hot metal. The committee machine was developed to model the hot metal during the operation of a coke blast furnace, according to the input parameters provided. The results obtained by the committee machine were lower than those of the neural networks acting alone, and the following RMSE values were verified: i) fuel rate: 4.88 (network 1), 4.74 (network 2), 6.14 (network 3) and 4.67 (committee); ii) sulfur content: 0.00915 (network 1), 0.00917 (network 2), 0.00974 (network 3) and 0.00726 (committee). Considering the results obtained, the model can be used to provide important support in monitoring and decision making during the operation.