Abstract: Profile roll bending forming is suitable for the production of high-precision and complex curved products in fields such as aerospace.During the progressive dynamic forming of profile, the forming moment and stress changes at different times have complex coupled and associated influences on the forming state, making it difficult for existing methods to accurately predict and effectively control the final forming state of profile roll bending.Therefore, establishing a roll bending forming prediction model and studying the dynamic forming laws of profile roll bending are key issues that need to be urgently solved in the precision forming industry.An intelligent prediction model for roll bending forming based on CNN(Convolutional Neural Network)-GRU(Gated Recurrent Unit)-KAN(Kolmogorov-Arnold Network) is proposed in this paper.Firstly, the progressive dynamic forming working principle of profile roll bending is analyzed.The CNN-GRU module is used to extract multi-scale correlation features of adjacent forming states, and an adaptive weighting method is adopted for effective feature fusion.On this basis, the nonlinear fitting ability of KAN is utilized to analyze the nonlinear influence of stress accumulation during the profile roll bending forming process on the final forming, achieving accurate prediction of roll bending forming.Experiments prove that the CNN-GRU-KAN model has higher prediction accuracy compared to other prediction models, providing an effective solution for the precision forming field.