Face recognition technology is increasingly applied in various fields such as commerce, security, and identity verification, existing single models struggle to cope with complex and variable real-world scenarios. Therefore, this paper aims to develop an artificial intelligence face recognition system based on a multi-layer weighted ensemble algorithm to improve the accuracy and robustness of recognition. This paper significantly improved the robustness and generalization ability of the AI face recognition system by introducing a multi-layer weighted integration algorithm. The main achievements of the project include new feature extraction methods (such as Local Mean Pattern LMP, Symmetric Local Graph Structure Descriptor SLGS, and its improved version V-SLGS), multi-feature fusion algorithms (grade score fusion decision-making and global-local double weighted integration), dynamic weighted integration strategy, and deep collaborative training and integrated decision-making methods. These innovations have performed prominently in improving recognition accuracy, enhancing system robustness, and optimizing real-time performance.