强化学习在游戏、机器人领域取得巨大成功，在CFD领域同样展现出可观潜力，帮助解决流体中机器人控制、外形优化设计等难题。然而，结合强化学习与CFD需要多领域知识，开发难度大，标准通用的开发框架的缺乏阻碍了相关应用发展。.本项目借鉴当前强化学习模拟环境和开源CFD软件架构，设计面向强化学习的通用可扩展CFD开发框架，实现与主流强化学习算法库的无缝连接，提供可扩展的开发接口；基于强化学习多幕训练特点，突破初值优化、精度hp自适应、并行任务调度等关键技术，提升训练效率；将框架及方法进行典型案例验证与新应用开发，并在国际平台上进行开源推广。本项目将是我国CFD软件自主创新的一次积极尝试，同时也助力强化学习突破现实鸿沟，具有显著的科研与工程实践意义。

Reinforcement learning (RL) has gained great success in the field of gaming and robotics. RL has also shown considerable potential in Computational Fluid Dynamics (CFD), helping to solve problems such as robot control and shape design in fluids. However, combining RL with CFD requires muti-disciplinary knowledge. The lack of a general development framework hinders the development of related applications..Inspired by the current RL simulation environment and open-source CFD software architecture, this project designs a general scalable CFD development framework for RL. The framework enables a seamless connection with the mainstream RL algorithm libraries and provides a scalable development interface. Considering characteristics of RL multi-step training, we propose using initial value optimization, hp adaptation and parallel task scheduling to improve training efficiency. The framework and methods will be verified on typical benchmarks. We will also try to spread the framework for new applications. The final project will be shared on the international open-source platform. This project will be a significant attempt at China's CFD software innovation. Also, this work can help RL to break the real-world gap. The outcomes are significant to scientific research and engineering practice.