Occluded person re-identification faces challenges such as missing key information, increased feature noise, and semantic misalignment under complex occlusion scenarios. Existing methods often suffer from strong dependence on external information and insufficient realism in data augmentation. To address these issues, this paper proposes a novel occluded person re-identification method based on realistic occlusion simulation and mask-guided suppression. Specifically, it proposes an occlusion image dataset with irregular boundaries and realistic textures, combined with a position-guided mechanism to effectively simulate the spatial distribution of real-world occlusions. Meanwhile, pixel-level occlusion masks are automatically generated and used to guide a dynamic suppression module that helps the Transformer network adaptively adjust its attention distribution. This targeted suppression reduces the impact of occluded regions on feature learning, significantly improving recognition accuracy and robustness under complex occlusion conditions. Experimental results demonstrate that our method achieves an average improvement of 5.6% in mAP over existing approaches, verifying the effectiveness of the proposed strategy. The occlusion image dataset will be publicly available through a GitHub repository.