With the rapid development of intelligent manufacturing, modular smart factories have become the key to improving production flexibility and efficiency. However, traditional factories face problems such as high reconstruction costs and slow response speed when responding to market changes and production demand adjustments. This paper proposes a dynamic reconstruction algorithm for modular smart factories based on digital twins. By building a virtual factory model, it realizes real-time interaction and synchronous optimization between the physical factory and the digital model. The study first analyzed the reconstruction needs and challenges of modular smart factories, and then designed a three-stage algorithm framework including dynamic demand prediction, module configuration optimization, and reconstruction path planning. Experimental results show that the algorithm can significantly shorten the reconstruction cycle. The space utilization rate before reconstruction is 72.0%, indicating that a lot of space is not fully utilized. After algorithm reconstruction, it increases to 85.0%, more space is reasonably utilized, reducing resource waste and enhancing the factory’s ability to adapt to market changes.