Abstract: Metasurfaces have demonstrated significant value across various fields due to its exceptional ability to manipulate electromagnetic waves. However, the highly complex nonlinear mapping relationship between their geometric configurations and electromagnetic responses has led traditional metasurface design methods to rely on massive electromagnetic simulations, severely constraining design efficiency. To achieve a breakthrough in this bottleneck, this paper proposes a fully automated design platform based on a high-performance phase prediction model using deep neural networks for the rapid design of focusing metasurfaces.This study generated a uniform dataset of 20,000 sets of geometric data–reflection phase pairs across an ultra-wideband frequency range of 2 to 18 GHz. A KAN-CNN（Kolmogorov-Arnold -Convolutional Neural Network）module, attention mechanisms, and residual connections—are utilized to construct a phase prediction network. This network enables the rapid generation of metasurface structural parameters from target phase profiles.Experimental results demonstrate that the platform achieves a high prediction accuracy of 92.7% for broadband reflection phases, significantly enhancing the efficiency of the design workflow. The platform ultimately designs a high-performance focusing metasurface array with a focal length of 100 mm operating in the 8 GHz band.