Prediction of the mechanical properties of hot rolling products by using attribute reduction ELM

Abstract

Aiming at the process quality parameters of hot rolling production with high-dimensional, strong coupling and redundant information and other features, a new mechanical properties prediction model is proposed. The model based on ELM algorithm is combined with the attribute reduction method. Firstly, attribute reduction method of combining information entropy with Gram-Schmidt orthogonal transformation is used to select effective process quality parameters and form the feature subset. And then, mechanical properties prediction model is built through adopting ELM algorithm as a neural network training approach. Finally, the model is proved by actual production data of two different hot rolling products from a certain iron and steel company. Compared with the traditional modeling methods, the model has the advantage of a simple structure, is less time-consuming, has high prediction accuracy, etc. The prediction results show that it is more adaptive to the complicated hot rolling process and the prediction performance is superior to the classic ELM model.