Tuning synchronization and memory characteristics of unsteady wake flows for physical reservoir computing

The performance of physical reservoir computers (PRC) depends heavily on the synchronization and memory characteristics of the chosen physical nonlinear system and the design of input and output layers. Here, we employ a periodic cylinder wake as an spatial distributed system for PRC and explore physics-informed strategies to tune the synchronization and memory characteristics that influence its PRC performance. An forcing that generates signal-encoded vorticity upstream of the cylinder is designed to effectively trigger the response of the wake flow. We observe that the synchronization of the wake to the input signal can be enhanced by appropriate choices of the frequency bandwidth and amplitude of the forcing, which greatly improves the PRC performance. In addition, an upstream-sensor-informed rolling of the cylinder can also improves the PRC performance due to faster wake synchronization and enhanced memory that preserves the high frequency components in the forcing. Furthermore, the memory capacity can also be addressed by the streamwise extent of the wake window as the subset of reservoir states in the output layer, since it preserves convective physics over a longer time horizon. These findings demonstrate a physics-based design for the input and output layers of a flow-based PRC for its tunable performance.