As global warming intensifies, extreme weather events become more frequent and severe, exacerbating the challenges of flood prevention. Hydraulic models, as core algorithms in digital twin technology, provide critical dynamic and real-time flood inundation information for flood prevention decision-making, holding immense practical application value. However, when it comes to large-scale flood prediction based on hydraulic simulation, the dramatic increase in computational demand leads to longer simulation times, affecting timely and effective flood prevention decisions. To overcome this challenge, the adoption of parallel acceleration computing techniques in hydraulic models has become an effective solution.On the other hand, from the perspective of national security, an important strategic goal of our country's technological development is to achieve domestic substitution of software in the water conservancy field. Through continuous independent research and innovation, China's water conservancy software industry has gradually reduced its dependence on foreign software, moving towards self-reliance and control. Domestic software has shown significant advantages in protecting China's water conservancy data security and preventing data leakage; at the same time, the promotion and use of domestic software can help reduce the cost of water conservancy informatization construction and improve investment benefits.

This paper elaborates on the high-performance two-dimensional hydrodynamic model developed by our team based on the domestic Taichi programming language—SkyFRAS, SkyFRAS possesses flexible parallel computing capabilities, is compatible with both homogeneous CPU-CPU and heterogeneous CPU-GPU computing architectures and supports a variety of parallel acceleration technologies including multi-core CPUs, CUDA, OpenGL, Metal, and Vulkan. Moreover, the model can operate smoothly on Nvidia GPUs as well as domestic GPUs, demonstrating its exceptional cross-platform performance. This research employs an unstructured grid to discretize two-dimensional space and utilizes the finite volume method for the numerical discretization of the continuity equation and momentum equation. By applying the SkyFRAS model to the flood calculation of the Raoyang River in Liaoning Province, it was found that compared to the original serial code, parallel processing achieved an acceleration ratio of up to 14.7 times.