Use of adaptive weighted echo state network ensemble for construction of prediction intervals and prediction reliability of silicon content in ironmaking process

The silicon content of molten iron is one of the most important molten iron quality parameters. However, the silicon content cannot be measured directly, therefore, accurate prediction for silicon content is of great significant to blast furnace (BF) iron making process. Aiming at the problem of low accuracy, an adaptive weighted echo state network (AW-ESN) based ensemble model is proposed in this paper to construct the prediction intervals (PI) and predict the silicon content of molten iron in BF. First, bootstrap method is utilized to resample the training set to construct subsets, AW-ESN is proposed to estimate silicon content and the corresponding PI is constructed. Then, the correspondence between the width of PI and reliability is established. Finally, the prediction results and the reliability can be obtained. In order to verify the effectiveness of the proposed method, industrial experiments were carried out by using process data of BF. The results demonstrate that the proposed method has higher prediction accuracy and the reliability can be realized, which provide more information to the on-site operators.