Abstract: In the production of non-ferrous metal plates, strips, and foils, the shape control system has nonlinear, multivariable, and time-delay characteristics, making it difficult to achieve satisfactory control results through conventional control methods. Based on the grey residual error and CNN-BiLSTM(Convolutional Neural Networks- Bidirectional Long Short-Term Memory) model, this paper proposes a predictive rolling optimization control algorithm for accurately predicting the tensile stress value of plate shapes between roll gaps. With the parallel computing capability and error compensation technology of CNN-BiLSTM network, the proposed algorithm overcomes the shortcomings of no feedback mechanism and inaccurate dynamic process prediction value in residual prediction. As a result, it obtains more accurate full-process prediction data in rolling and compensates for the inertia and lag characteristics of the plate shape control actuator through rolling optimization control. The prediction accuracy of the grey neural network combination algorithm is compared with that of various models such as grey prediction, BP, CNN, and LSTM. The prediction accuracy is increased by 7.46%, 6.28%, 17.85%, and 17.2%, respectively, and the dynamic control accuracy is improved by 30%. This proves the effectiveness of the proposed algorithm and provides a new avenue for shape control and real-time data prediction.

Key words:  , Flatness control , Grey residual prediction model , CNN-BiLSTM , Rolling optimization