Application of Recurrent Neural Networks to Wind Speed Prediction from Flapping Flags

Neural networks have shown great promise in their ability to model complex dynamic systems. Here, recurrent neural networks are applied to time series data from video clips of flapping flags in order to predict wind speeds. Details of the trained models are examined alongside observed flag kinematics to determine which physical aspects of the motion are important for the model to make accurate wind speed predictions. Network activations are analyzed to extract salient physics of the system.