The Research on Finish Rolling Temperature Prediction Based on Deep Belief Network

Abstract

A method based on deep belief network (DBN) is proposed in this paper to improve the accuracy of finish rolling temperature prediction in the finish rolling temperature control system. DBN is composed of a plurality of restricted Boltzmann machines (RBM) and a top-level BP neural network. Taking into account the factors affecting the finish rolling temperature and the practical production requirements, 10 input layer parameters are set in this model, and the output layer parameter is the finish rolling temperature. Unsupervised training for restricted Boltzmann machines and the reversed fine-tuning of the entire network is obtained by 1300 sets of finishing data. After simulation, the absolute error fluctuation range of the predicted temperature is less than 8°C, and its prediction accuracy is higher than that obtained from the traditional temperature calculation formula, thus the proposed method can be used for the finish rolling temperature prediction.