Abstract: The crown of cold-rolled non-oriented silicon steel strip is the result of the coupling effect of process parameters such as roll shifting, roll bending force and rolling force. It is difficult for traditional mechanistic models to predict and optimize the strip crown timely and accurately in production. To solve this problem, based on Deep Neural Network (DNN) and Whale Optimization Algorithm (WOA), this paper proposes a model for crown prediction and optimization of roll bending force and roll shifting in the cold rolling process of silicon steel. Taking variables such as the width and crown of incoming cold-rolled strip, roll shifting and roll bending force of each stand as inputs, the C₂₅ value of the cold-rolled outlet strip is predicted. Guided by the prediction results of the data-driven model, WOA is used to optimize roll bending force and roll shifting, and Latin Hypercube Sampling (LHS) is introduced to accelerate the optimization. The results show that the model can achieve high-precision prediction of C₂₅ crown of silicon steel strip, with a hit rate of 99.5% for errors within 0.5 μm. By optimizing roll bending force and roll shifting, the C₂₅ crown of cold-rolled silicon steel strip is reduced by 0.89 μm, and the slope of the edge drop region is decreased by 13.9% at the same time, solving the problem of excessive edge drop in cold-rolled silicon steel.