Research on rolling force modelling and interpretability of cold rolling based on industrial big data

In cold rolling production, rolling force is the decisive parameter that causes the strip to undergo plastic deformation, and its setting accuracy directly affects the thickness accuracy, plate shape quality and edge drop control effect of products. However, traditional rolling force models rely on a large number of simplified assumptions, resulting in limited calculation accuracy. To overcome the above problems, this study proposes a rolling force prediction and interpretable method based on industrial big data for cold rolling. This method first develops a cross-process industrial IoT platform to integrate key process data of hot-cold rolling lines and provide a complete industrial dataset for model training. Then, an adaptive inertia weight strategy is introduced to improve the whale optimization algorithm (IWOA), and the hyperparameters of the stochastic configuration network (SCN) are optimized to construct a rolling force prediction model. Finally, the SHAP method is used to analyse the characteristic contribution of each input feature of the hot-cold rolling line to the rolling force prediction results, thereby enhancing the interpretability of the rolling force prediction process. The experimental results demonstrate that compared with baseline models or other advanced models in the rolling field, the proposed IWOA-SCN exhibits the best prediction accuracy and generalization ability in rolling force prediction. In industrial experiments, the proposed model significantly reduces the rolling force prediction deviation, further verifying the accuracy and adaptability of the method under complex working conditions, and providing strong support for achieving precise process control and improving product quality in cold rolling production.