Mechanical Engineering Commons^™

Deep Learning Of Nonlinear Dynamical System, Aditya Wagh

Dissertations, Master's Theses and Master's Reports

A data-driven approach, such as neural networks, is an alternative to traditional parametric-model methods for nonlinear system identification. Recently, long Short- Term Memory (LSTM) neural networks have been studied to model nonlinear dynamical systems. However, many of these contributions are made considering that the input to the system is known or measurable, which often may not be the case. This thesis presents a method based on LSTM for output-only modeling, identification, and prediction of nonlinear systems. A numerical study is performed and discussed on Duffing systems with various cubic nonlinearity.