Data-driven reduced order control for partially observed fluid systems

We propose a novel method for non-intrusive reduced order control of partially observed flow systems. We formulate a rank-constrained matrix optimization problem for the maximum likelihood estimation of the reduced order model. An adjoint-based method is used for the gradient extraction and Riemannian optimization is performed for efficient convergence to the optimal solution. The resulting reduced order model is then used to design a Linear-Quadratic-Gaussian (LQG) controller. We demonstrate the performance of the proposed reduced order control method on the flow past an inclined flat plate at a high angle of attack and successfully prevent vortex shedding in the wake of the flat plate.