Research On Dynamic Effective Matrix Flatness Control Based On Cloud Adaptive Differential Algorithm

With the increased requirements of high quality plate and strip products, flatnessshape control becomes the key technique and important develop direction for moderntimes high precise rolling mill. Recently the artificial intelligence theory which has a goodfunction on model setting, optimization and controlling was widely used in the field offlatness pattern recognition and flatness control system. Based on the theory of artificialintelligence, the embedded theoretical research on flatness prediction and controlintelligent method is presented in the paper.Firstly, an adaptive differential evolution algorithm based cloud-model is proposed.Because the traditional differential algorithm has the disadvantage of prematureconvergence and searching stagnation, the basic normal cloud generator is used toimprove the differential evolution algorithm and then to complete the variation of theCADE model. And the adaptive operators are introduced for the selection of digitalcharacteristics of cloud model during the process of producing new individuals. Moreover,the robustness, randomness and stable tendency of the algorithm are enhanced.Secondly, the flatness forecasting model based on CADE-BP network is presented.After preprocessing the practical statistical producing data, the forecasting model based onthe CADE-BP network is established by choosing the important influence factors offlatness during rolling process as the input, and the character coefficients of flatness curvefitting as the output. According to the trait of experiment, the global optimization ability ofCADE algorithm is used to optimize the BP neural network’s weights and thresholds toimprove the prediction efficiency and precision of the flatness forecasting model.Thirdly, the transfer effective matrix flatness control model based on the CADE-BPnetwork is presented. On the basis of the predicting model of flatness control, the dynamicmatrix method of flatness control is proposed. The empirical value table of effectivematrix is produced by ascertaining the key effective factors of flatness control mechanism.The real-time feature has been considered to adjust the effective matrix and provide theflatness control with accurate information.At last, the simulation experiments of the flatness forecasting model and the flatness controlling model are finished by using Matlab software.