基于多模态大语言模型智能体的小流域设计洪水推求研究OA

Design flood estimation for small watersheds is crucial for flood control and disaster mitigation.Tradi-tional methods suffer from issues such as reliance on manual chart lookup for parameter acquisition,rigid calcula-tion procedures,and low levels of intelligence,making them inadequate to meet the demands of modern water con-servancy construction.To address these issues,this paper proposes an automated design flood estimation method for small watersheds based on a multimodal large language model(MLLM)agent.First,a multimodal design flood parameter dataset integrating spatially interpolated raster data and digitized charts and tables is constructed using data from discrete hydrological stations in Guangdong Province and the User Manual of Storm Runoff Calculation Charts and Tables for Guangdong Province.Second,an agent centered on DeepSeek-V3 is developed,around which a modular API system is designed,encompassing data services,computation services based on the rational formula method,and computation services based on the Guangdong Province unit hydrograph method.Finally,through a tool-calling mechanism,the agent autonomously understands natural language instructions,performs task planning,and invokes tools to accomplish the entire automated workflow from data acquisition to report generation.The pro-posed method is validated using three typical ungauged small watersheds within the Pajiang River Basin as case studies.The results demonstrate that the agent can autonomously complete the entire workflow from task parsing to report generation,and the computed results are highly consistent with those produced by the Guangdong Hydrologi-cal and Hydraulic Design and Computing Software Platform.The proposed method provides a new technical path-way for automating complex hydrological calculation processes,verifies the effectiveness and accuracy of MLLM agents in the field of smart water conservancy,and demonstrates broad application prospects in flood control plan-ning for ungauged regions and the construction of digital twin basins.