Flatness prediction of plate and strip straightening based on GA-CNN model

To address problems such as that the traditional design of straightening parameters relies on experience and the prediction model suffers from insufficient accuracy, this paper proposes a GA-CNN hybrid model integrating genetic algorithm (GA) and convolutional neural network (CNN). Based on the data from a domestic heavy plate production line, a high-quality dataset is constructed through standardization, five-fold cross-validation for five times and data folding, and a two-stage optimization strategy is innovatively adopted. In the first stage, a CNN model is built to capture plate flatness distortion features; in the second stage, GA is introduced to adaptively optimize network structure and hyperparameters, improving the model prediction accuracy by 26.4%. The results show that the key indicators of the GA-CNN model are significantly better than models such as DNN and random forest, with its mean absolute error of 1.15 and root mean square error of 1.91. In addition, its accuracy on the training set and test set is the highest among other algorithmic models, reaching 94.73% and 93.06% respectively. This paper not only provides theoretical support for intelligent decision-making of straightening process parameters, but also promotes the paradigm transformation of plate and strip manufacturing from experience-driven to data-driven, and provides a reusable technical path for the intelligent manufacturing upgrading of the iron and steel industry.