The work rolls of CVC rolling mills in plate rolling process adopt S-shaped roll profile curve, which change the contact situation between the rolls and make the contact pressure distribution between the rolls uneven. The peak of the contact pressure between the rolls appears at the end of the roll, which leads to peeling or “meat loss” at the end of the roll and poor plate crown control effect. To solve those problems, a CVC rolling mill back-up roll profile curve with parabolic chamfer was proposed, and Fusion model of plate crown based on mechanism and data-driven models was developed. Combining the parameters of the CVC roll profile curve, the theoretical calculation formula of the back-up roll chamfering was deduced. According to the actual data of the plate rolling mill, the parameters of different types of back-up roll chamfering were determined, and the distribution of contact pressure between the rolls under different back-up roll chamfering was calculated and analyzed by finite element model. Based on the new parabolic chamfer curve of back-up roll, the rolling force setting model based on ELM and the roll wear model based on GA-ANN were constructed by using the nonlinear fitting ability of machine learning algorithm, Then, the predicted results of rolling force and roll wear are taken as the input variables of plate crown prediction. At the same time, the roll system deformation model and roll bending force model based on mechanism model were introduced as inputs to construct the plate crown CNN model based on mechanism and data-driven fusion. The simulation results show that the parabolic chamfer can significantly reduce the pressure between the backup rolls. When the rolled strip width is 2000mm, 2400mm, and 2800mm, the parabolic chamfer can reduce the peak contact stress between the rolls by approximately 8.6% and 10.8%. When the rolling bending force is 1600kN, 2100kN, and 2600kN, the parabolic chamfer can reduce the peak contact stress between the rolls by approximately 8.6%, 8.9% and 9.1%. The hit ratio of plate crown increased from 90.4–96.5%, and the product degradation rate caused by the shape problem decreased by 35.5%.