Abstract: To address bottlenecks in current water cycle simulation technologies, such as low efficiency, strong subjectivity, and "knowledge silos" when confronting complex water security challenges, we proposed a "human-computer collaborative" framework for water cycle process simulation that is driven by a large cognitive model of water conservancy and guided by a domain knowledge base, in line with the strategic needs for the development of digital twin basins. This framework aims to advance the water cycle simulation paradigm from "model-centric with a human operator" to "cognitive intelligence-centric with a human supervisor." It achieves closed-loop optimization of the workflow through three core stages. First, in the modeling stage, it depends on demand-driven dynamic modeling technology to realize rapid comprehension and on-demand modeling of complex hydrological scenarios. Second, during the simulation stage, it utilizes online adaptive simulation technology to enable real-time optimization and dynamic adjustment of forecasting and decision-making processes. Finally, in the post-event review stage, it promotes continuous iteration and long-term learning for the large model and knowledge base through human-computer collaborative event review and knowledge extraction. An application case demonstrates that the framework can effectively digitize the tacit knowledge of experts by transforming the manual, expert-dependent workflow of traditional forecasting—involving topology construction, scenario modeling, model adaptation and parameter optimization—into an intelligent, automated online optimization process, thereby significantly enhancing the responsiveness and decision-making efficiency for flood events. This research offers a new theoretical framework and technical pathway for tackling the challenges of knowledge inheritance in the water conservancy sector and advancing intelligence of water cycle simulation, providing significant support for the construction of high-order smart water conservancy systems.