| Plate and strip play a important role in the industrial production and people's life, so the improvements of their production quality become the focus of academic circles. Traditional control theory can't meet the modern industry requirement of product precision very well, but the improvement and development of artificial intelligence theory, especially the artificial neural network theory, provide a more powerful tool for the production of plate and strip. In recent decades, artificial neural network has been continuously used in flatness pattern recognition and control. Faced with the problem of flatness intelligent recognition and control, a flatness recognition system and control system based on Elman network are designed, and a further research on neural network training methods is made.Firstly, because the ability of generalization of BP network is poor, its learning speed is slow, a flatness pattern recognition model based on Elman network is established. The results show that the abilities of learning and generalization of the model based on Elman network are better than the model based on BP network with the same number of the hidden node and the epoch of training, so the Elman network gets good results.Secondly, prediction model for flatness control is established based on Elman network trained by several kinds of algorithm, such as error back propagation algorithm, particle swarm optimization algorithm, simulated annealing algorithm, Artificial Bee Colony algorithm and bayesian regularization algorithm. The model based on Elman network trained by bayesian regularization algorithm is adopted based on training time, the ability of generalization and some other norms. Dynamic effective matrix for flatness control is gotten from the model based on Elman network trained by bayesian regularization algorithm. Finally, dynamic effective matrix for flatness control is applied to 900HC rolling mill. This method achieves good results that is proved by simulation. |
PDF Full Text Request