Extreme heavy rainfall (EHR) events pose severe threats to lives and property worldwide, and their frequency and intensity are escalating with global climate change. Nevertheless, our understanding of the physical mechanisms governing these events remains markedly limited. Recently, artificial intelligence (AI) models have increasingly come to rely on well-documented individual EHR case data to capture the diverse precursor pathways and organizational modes that drive extreme rainfall. At the heart of this knowledge gap lies the observational limitation in capturing the full life-cycle evolution of severe convective systems. Although atmospheric reanalysis datasets are widely used in EHR research, they fall short of adequately representing EHR. To address this deficiency, we propose the Duoyu Project, which employs the Weather Research and Forecasting (WRF) model to reconstruct EHR episodes and generate a comprehensive suite of high-resolution numerical simulations, thereby providing a robust data foundation for advancing EHR research. To date, we have completed simulations for 10 EHR cases over China, using a horizontal resolution approximately 1 km with 58 sigma levels. The standard output interval is 6 minutes, with 1-minute resolution for selected cases. The simulated data have been evaluated against surface observations and radar data, and overall show good agreement with observations, capturing both spatial patterns and temporal evolutions. Basic information and updates are available on the homepage of the Duoyu Project (https://www.rainstormproj.com/) and also on this Zenodo repository. Sample data are stored here. All the data can be obtained as a hard disk copy by contacting the authors at present, and will be freely accessible at this location (https://www.rainstormproj.com/) in the near future.