Research On Model Predictive Control Of Cold Continuous Rolling Plate Shape Based On Data Driven

This paper implements a method of predictive control with a steel plate shape(Model Predictive Control,MPC)by taking the 1450mm five-stand UCM cold continuous rolling production line of a steel plant as the research object,and taking the identification method of cold rolling plate shape control efficiency coefficient and the prediction model of cold rolled plate shape as the starting points.Compared with the traditional pattern recognition or finite element method,this paper proposed a partial least squares algorithm integrated with deep learning to obtain the control efficiency coefficient of the shape from the perspective of data.Compared with the traditional single-input control strategy that sequentially eliminates the remaining deviations of the plate shape according to the priority of the control structure,a cold-rolled shape model predictive control system was designed to realize the optimal control of the shape.The main research contents of this paper are as follows:(1)Firstly,the production data of the history of cold rolling is extracted from the database,and the data is preprocessed.According to the characteristics of nonlinearity,strong coupling and multi-noise of cold rolling data,combined with the influence law of various control mechanisms of the plate shape,the partial least squares method is used to obtain the efficiency coefficient of the plate shape regulation.The partial least squares algorithm is improved for the nonlinear characteristics of the data,and the sparse autoencoder model is integrated to improve the ability to extract nonlinear features in the data.The effects of the control partial-minimum algorithm and the improved partial least squares algorithm on the control efficiency coefficient are compared by experiments.The accuracy of the latter’s control efficiency coefficient is verified,and the efficacy coefficient with good effect can be obtained.(2)Considering the related factors affecting the shape of cold-rolled sheet,the SAE-PLS+ANN algorithm which can realize nonlinear predictive regression was proposed,and the cold-rolled shape prediction model was established.The sparse auto-encoder and the partial least squares algorithm are combined as the main predictive model,and then the neural network is added to train the prediction bias of the main model to compensate for a part of the prediction bias of the main model.Finally,the model is trained according to the actual production data of cold rolling,and the high-precision prediction of the shape value of the test steel coil is realized,and the prediction performance of the model is verified.(3)Based on the previous two steps,the MPC model predictive control method was established.The prediction model predicts the shape deviation that will occur at the set value in the future,and uses the gradient descent method to find the optimal control parameters when the remaining shape deviation is the smallest.The prediction is performed cyclically to achieve the rolling optimization of the shape.The scrolling optimization of the shape is realized by the circular shape prediction.By comparing the optimization effect of MPC dynamic optimization strategy and priority control strategy on the target shape,it is verified that the MPC method has better optimization effect under the synergy of multiple adjustment variables,and the root mean square error and the mean absolute error of the optimization method were reduced from 2.43 IU and 1.72 IU to 1.95 IU and 1.36 IU,respectively.