Prediction of hot strip width based on rolling mechanism and hybrid neural network
WANG Xiao-wen1, ZHANG Yong-jun1,2, GUO Qiang1,2, ZHANG Fei1,2, PEI Hong-ping2
1. Institute of Engineering Technology, University of Science and Technology Beijing, Beijing 100083, China; 2. National Engineering Research Center for Advanced Rolling and Intelligent Manufacturing, University of Science and Technology Beijing, Beijing 100083, China
Abstract:The width accuracy of hot-rolled strip products directly affects the yield of products and is the key to improving product performance. The accurate prediction of the strip exit width in the finish rolling area can provide timely optimization and adjustment guidance for the width control model parameters in the rough rolling area. The traditional mechanism model is often different from the actual situation. Most of the existing data-driven models use neural network method,but do not consider the timing of rolling data and the information loss caused by data pruning. In order to further improve the precision of width prediction for finish rolling strip,a hybrid neural network width prediction model based on rolling mechanism was proposed. The width reference value was calculated using the mechanism model of finish rolling spread,and the width prediction correction value was output by combining convolutional neural network (CNN) and gated recurrent unit (GRU). The data set test of a 2 250 mm hot continuous rolling mill shows that the training efficiency of the proposed model is high, and deviations for 98.7% testing samples are within 4 mm, which is greatly improved compared with the traditional BP neural network model and other single structure networks, and the on-line test speed of the model meets the industrial application requirements.
王晓雯, 张勇军, 郭强, 张飞, 裴红平. 基于轧制机理和混合神经网络的热轧精轧带宽预测[J]. 中国冶金, 2023, 33(2): 114-120.
WANG Xiao-wen, ZHANG Yong-jun, GUO Qiang, ZHANG Fei, PEI Hong-ping. Prediction of hot strip width based on rolling mechanism and hybrid neural network[J]. China Metallurgy, 2023, 33(2): 114-120.