An integrated model of rolling force for extra-thick plate by combining theoretical model and neural network model

To solve the problem of low precision of the existing theoretical model in predicting the rolling force of extrathick plate, the genetic algorithm (GA) is used as an enhancing means to improve the global searching ability of the BP model, and a GA-BP model with high precision is firstly established. Furthermore, in order to solve the black box problem of the model, an integrated model is ultimately obtained by combining a theoretical model and the established neural network model. During the modeling, 1000 groups of production data of extra-thick plate rolling are selected and normalized as the data set. The optimal network structure of the GA-BP neural network is determined based on the method of trial and error, and the initial weight and threshold of the BP neural network is solved iteratively with the genetic algorithm. On this basis, an integrated model is ultimately obtained according to the principle of multiplication compensation of average error. It is shown that the maximum prediction error of the original BP model is 7.51%, while the value of the GA-BP model is down to 3.95%. This integrated model has not only inherited the rigorous mathematical structure of the theoretical model, but also occupies the high precision that comes from the GA-BP model. Therefore, the present integrated model is more suitable for the process optimization of extra-thick plate rolling.