Flow modeling and control employing dynamic mode decomposition and deep reinforcement learning

The high dimensionality of fluid flow simulations poses significant challenges in terms of computationalcost and memory usage, motivating the development of reduced-order modeling techniques.Dynamic Mode Decomposition with Control (DMDc) is one such technique that provides a frameworkfor approximating the time evolution of the flow state while accounting for flow actuation.In this work, DMDc is used as a surrogate model to accelerate the optimization of control laws inclosed-loop active flow control applications using deep reinforcement learning. To address memorylimitations in large-scale flow problems, a partitioned singular value decomposition algorithmis employed to enable memory-efficient DMDc model construction. Overall, this project aims toreduce the computational cost and turnaround time of flow simulations while alleviating memoryconstraints in data-driven flow modeling and control.