高速运动体流动减阻与主动流动控制密切相关。本项目针对视觉测量的光流场计算，研究融合流动机理的变分优化计算；探讨流动机理与数据驱动的混合深度学习新方法；基于流动机理PDE模型，通过能观性度量及约束Lyapunov方程，建立传感器位置优化方法；引入具有非局部边界条件的线性化Navier-Stokes方程以及对应的涡量-流函数PDE，分析边界反馈参数能稳性范围。以圆柱绕流为例，拟建立边界反馈控制参数与物理参数（如雷诺数等）间的解析依赖关系；运用深度学习，建立控制到流场之间的映射网络研究数据驱动的反馈参数学习，避免机理模型与实际情况的失配问题；基于浙大水洞实验装置，设计搭建可转动圆柱绕流实验，验证图像测速、传感器优化部署、视觉反馈的主动控制等理论结果。项目特色是提出混合学习（Hybrid Learning）方法，研究流动场视觉测量与传感位置优化、光流场计算与同化、圆柱旋转反馈控制器设计及实验验证。

Modeling and active control of fluid flows are much related to the drag reduction of high-speed vehicles. In this project, physics-constrained optical flow computing is further explored via variational optimization toward visual measurements. Hybrid deep machine learning framework is proposed to combine fluid physics and data-driven approaches in a synergic framework. Based on the first principle PDE models, observability metric is derived as well as its governing operator Lyapunov equation, and then a PDE-constrained optimization problem is established for sensor deployment. For the 2D flow past a rotary cylinder benchmark, the point and local feedback control is considered where the PDE models have non-local boundary conditions. Then, stability analysis in terms of parameters in the feedback form is carried out towards an explicit dependence of fluid parameters, such as cylinder diameter, Reynolds number, etc. Deep learning with a constraint from fluid physics, or hybrid learning is proposed in this project to solve the visual computing, sensor deployment as well as control parameter tuning. The hybrid learning could probably increase the learning efficiency while overcoming the modeling uncertainties in the reality based on the first principle. Finally, experiments are settled up based on the existing hydraulic tunnel in Zhejiang University to validate all the theoretical results.