Hybrid Finite Gain Stable Network to Identify Ship Motion in Open Water

This thesis incorporates prior knowledge in the form of linearization of an unknown nonlinear differential equation with a robust recurrent neural network to make predictions of the system states.

Running Master Thesis


Modeling and identification of dynamical systems play a crucial role in simulation and control engineering, the goal is to get an accurate representation of the unknown system. Compared to classical modeling techniques, which require expert knowledge, deep learning approaches are able to learn the dynamics purely from measured input-output data. However, deep learning in the form of sequence-to-sequence models (e.g. recurrent neural networks) lacks proof of stability and robustness often needed in classical engineering. In this work, our goal is to combine classical modeling techniques, where we assume to have some prior knowledge of the system, with a recurrent neural network that has learnable weights. The recurrent neural network has constraints on the parameters to get a guaranteed upper stability gain. As prior knowledge, we assume to know a linear approximation of the nonlinear ship dynamics. By interconnecting the two models, where the constrained recurrent neural network learns the residual of the linear, we hope to achieve higher prediction accuracy than purely learning-based methods and give guarantees on the input-output behavior of the hybrid system. We will evaluate our hybrid model in terms of stability as well as prediction accuracy against other learning-based approaches.


To the top of the page