RBF Neural Network‐Based Distributed Nonlinear Model Predictive Control on Tandem Cold Rolling Stands

ABSTRACT The precision of the strip thickness is an index of significance in the tandem cold rolling process which makes a difference to the strip quality. However, it is hard to build an accurate mathematical model for thickness control in the tandem cold rolling process, because there exist coupling relationships of complexity between the adjacent stands and unmeasurable process parameters. To overcome the difficulties, a distributed nonlinear model predictive control (DNMPC) strategy with a deep learning method is put forward in this paper, where the auto‐regressive radial basis function neural networks are established to model the tandem cold rolling process. For each stand, not only the control input and exit thickness output data of this stand, but also the data of the neighbor stands are selected as the input of the neural network. Besides, the design of the distributed nonlinear model predictive controller turns into an optimization problem, and the gradient method is applied to solve it, which gets rid of the leaning upon mathematical models. Moreover, the stability of the developed method is proven, which indicates the boundedness of the tracking error. The simulations are carried out on three‐stand and five‐stand examples, and the results verify the efficacy of the proposed DNMPC strategy.