Abstract: This study proposes a dynamic layout generation method integrating an agent-based reasoning mechanism to address the dynamic nature of shipyard block layouts, which must frequently adapt to docking schedule adjustments and process changes. By introducing agent technology into layout design, this approach enables an agile response to user requirements while precisely managing complex spatial constraints between equipment, demonstrating strong potential for adaptation and customization. To enhance the efficiency and accuracy of the model’s reasoning process, we constructed an agent-based framework for inference centered on task decomposition, planning, and evaluation-reflection. This framework connects multiple reasoning steps into a coherent chain, allowing the model to systematically analyze and solve complex layout problems in a phased, structured manner. This method not only mitigates information overload but also significantly improves the model's accuracy and efficiency in handling such complex tasks. Furthermore, the system establishes clear standards and decision rules to ensure the reliability of information processing and the validity of decisions throughout the multi-step reasoning iterations, guaranteeing the generation of effective inference sequences. Finally, a case study on the block storage yard of a large shipbuilding enterprise was conducted to validate the method's ability to reconfigure layouts in response to sudden docking schedule changes. The results show that the proposed method significantly improves the layout's responsiveness to dynamic demands and its constraint satisfaction rate while maintaining spatial utilization, offering a novel approach for the intelligent and dynamic adjustment of shipyard layouts.

Key words: LLM agent, digital twin, layout design, retrieval-augmented generation (RAG), shipbuilding