A prediction model for the width of medium thick plate based on the fusion of point prediction and interval prediction

The prediction of the width of medium thick plate is very important to increase the product yield. Most of the current research focuses on improving the prediction accuracy of the high point, but seldom considers the ability of the model to resist the risk, which is urgently needed in the existing hot rolling production system. In this paper, a fusion method based on point prediction and interval prediction is proposed, which can balance the accuracy and risk of the medium thick plate width prediction models. The main work is to screen out the original industrial process variables that have a great influence on the width of medium thick plate through hierarchical clustering, and to construct the point prediction and interval prediction models by combining the upper and lower bounds detection method of Deep Neural Network Model and Quantile Regression Neural Network. The experiments show that the proposed fusion of point prediction and interval prediction can take into account the accuracy and risk of medium thick plate width prediction model. The experiments show that the interval coverage ratio is 0.95, and the normalized average width of predicted intervals is 0.10, which indicates that the model has a very good prediction performance and risk resistance ability.