Study On System Of Cold Tandem Rolling Tension Using Neural Network And Predictive Control

AbstractIn this paper, after reading abundant relevant literature, domestic and oversea, we analyzed the mathematical model of tension in cold tandem rolling machine and the construction of control system and studied the theory of the generalized predictive control (GPC) and its design method. For the real time restriction of generalized predictive control in tension control of cold tandem rolling, we deduced a sort of quick arithmetic of generalized predictive control and presented a set of directive solution formulae of Diophantine equation and its demonstration that avoided the on-line recursive computation and decreased the burden of the generalized predictive controller in computation. Because tension control has the character of changing parameter, we proposed the idea that GPC combined neural networks. Actually, it used the neural networks as the identified model of control object to produce predictive signal, the control law was solved by optimized arithmetic. Accordingly we realized the predictive control of the changing parameter system. We choose BP and RBF neural networks as identified model for they could approach the function very well. First we identified the controlled object offline, when the precision reached a certain extent, We would achieve recursive predictive model by on-line identification. Finally we acquired the optimized control law by minimizing the function of performance guideline. This arithmetic not only improved the precision of model identification but also avoided on-line computation of Diophantine equation. Furthermore it simplified the calculation of matrix inversion, so the controller burden was decreased drastically. It also benefited its application to quick systems like cold tandem rolling, therefore, its application scope of predictive control was further broadened. In the end, we studied the generalized predictive pole-placement control which based on Hoptield neural network, the performance of quick optimized computation of neural network was introduced into predictive control, hence, it compensate the deficiency of predictive control and improved the real time character of arithmetic.