Flexible molds, with their excellent adaptability and reconfigurability, can meet the production needs of curved products with varying models, shapes, and sizes flexibly. In the design process of flexible molds, accurately determining the position and quantity of surface formation points is crucial. Therefore, this study innovatively introduces the construction of a Flexible Mold Surface Formation Neural Network model based on point cloud and neural network technologies, specifically for the extraction and determination of surface formation points in the design of flexible molds. Based on the TRIZ theory, we utilized the tools of local idealization and technical contradiction resolution to analyze the function and architecture of the Flexible Mold Surface Formation Network. And we analyzed the characteristics of the point cloud that are pertinent to flexible molds. Next, for the processing of the raw point cloud data, we introduced a lower edge point screening module based on the Support Vector Machine (SVM) and a filtering module. We then incorporated the PointNet++ network and made improvements to it, including the addition of a surface formation feature information integration module and the enhancement of the point cloud sampling method, thereby establishing a comprehensive neural network model dedicated to the surface formation of flexible molds. Finally, experimental data indicate that the method we have developed offers high precision and effectiveness, providing a practical theoretical foundation for the critical step of determining surface formation points in the design of flexible molds.