Data-driven prediction and control of extreme events in a chaotic flow
Alberto Racca, Luca Magri
Abstract
Extreme events are sudden and violent changes in the state of a nonlinear system, which may have adverse consequences on the system's components. We use recurrent neural networks to time accurately and statistically predict extreme events in a reduced-order model of chaotic shear flow at multiple Reynolds numbers. Through the predictions of the networks, we control the flow and reduce the occurrence of the events, therefore improving the operability of the system.
Topics & Concepts
OperabilityChaoticNonlinear systemFlow (mathematics)Computer scienceReynolds numberControl theory (sociology)Artificial neural networkControl (management)MeteorologyMechanicsArtificial intelligenceTurbulencePhysicsQuantum mechanicsSoftware engineeringModel Reduction and Neural NetworksNeural Networks and ApplicationsNeural Networks and Reservoir Computing