Generation of time response of linear and nonlinear dynamic systems using autoencoders

Abstract

In this work, a deep autoencoder is used to generate the time response of linear and nonlinear dynamic systems. First, the encoder part of the autoencoder is used to perform a compact representation of the time response of dynamic systems. Second, the decoder part of the autoencoder is used to reconstruct or generate the time response of dynamic systems from the latent space. Experiments are performed to determine the capability of the architecture and the training algorithm proposed for dimensionality reduction, reconstruction, and generation. Finally, the architecture is validated with some examples of linear and nonlinear systems.