A novel theoretical and data-driven hybrid model based on weight allocation strategy for predicting hot-rolled strip crown

Crown is the key indicator for evaluating the cross-sectional profile quality of hot-rolled strip. Accurate crown prediction can directly contribute to improving product quality and minimising potential risks. The theoretical model (TM) has a complete theoretical system, but its prediction accuracy is limited by the influence of simplified assumptions in modeling. The data-driven model (DDM) can ignore the cross-coupling influence between input features and has a high prediction accuracy, but the model lacks guidance from domain knowledge and is limited by the quality and scale of training data, making it prone to distortion. This study proposes a hybrid modeling approach that combines theoretical and DDMs. The method utilises the TM to calculate the crown theoretical value, after which it identifies the key features as inputs through the rolling mechanism guidance. Then, the modified ant colony optimisation (MACO) algorithm with a pheromone adaptation mechanism is used to optimise the input weights and biases of the extreme learning machine (ELM), thereby constructing the MACO-ELM DDM. Subsequently, the entropy weight method is used to assign weight coefficients to each model to correct the model results, and finally, the TM-MACO-ELM hybrid model is constructed. The experimental results show that this weight allocation strategy effectively reduces the impact of error transmission on model accuracy during hybrid modeling. Compared to single model or bias-compensated hybrid model, the TM-MACO-ELM model proposed in this study exhibits the highest accuracy and stability. Finally, a crown control strategy is proposed and used in industrial experiments, which effectively improves the level of crown control. The method proposed in this study can achieve high-precision crown prediction and control, which is of great significance for the improvement of product quality of hot-rolled strips.