Rolling force prediction based on wavelet multiple-RBF neural network

Rolling force prediction is very important in hot strip rolling mill. But during the setting of rolling force in hot strip rolling mill, the rolling force signal is influenced by various factors with complicated correlation. Therefore, it is difficult to establish an accurate model to describe the rolling mechanism. In order to solve this problem, a multiple-RBF neural network model to predict rolling force based on wavelet analysis is proposed. In the new model, the multiresolution wavelet analysis method is employed to separate the rolling force signal into several sub-signals corresponding to different factors, and several RBF neural networks are established, each for a certain subsignal. Thus, the sub-networks can well reflect the variation mechanism of the rolling force. With real-time data obtained from the production process, the proposed model is trained and applied to the rolling force prediction. Simulation results show that the prediction accuracy is improved, and the average error rate decreases to less than 5%. Therefore, the method is an effective way to build a prediction model of rolling force and has great potential for applications in practice with it's high prediction accuracy and short training time.