Off-campus SJSU users: To download campus access theses, please use the following link to log into our proxy server with your SJSU library user name and PIN.
Publication Date
Spring 2024
Degree Type
Thesis - Campus Access Only
Degree Name
Master of Science (MS)
Department
Computer Engineering
Advisor
Stas Tiomkin; Mahima Agumbe Suresh; Carlos Rojas
Abstract
This thesis proposes a new method to learn the behavior of hybrid systems from observations. A hybrid system is a type of dynamical system that exhibits both discrete and continuous behavior. In other words, it is a system where the state variables change depending on both continuous dynamics and discrete transitions. These two types of dynamics interact and produce intricate dynamical behaviors, such as switching when a continuous variable exceeds a threshold. Modeling and analyzing hybrid dynamical systems is challenging due to their nonlinear and discontinuous nature. Existing methods heavily rely on domain knowledge and assumptions and lack a unified approach for characterizing trajectories in hybrid systems. There are several mathematical representations of hybrid dynamical systems in the literature. One is the Differential-Algebraic-Discrete (DAD) model. In this thesis, we aim to parameterize the DAD model with neural networks and fit the model to data. This approach requires solving a multi-level optimization problem, where some variables are hidden, posing significant challenges. Our approach enables the prediction of system behavior with minimal reliance on domain knowledge and assumptions. We aim to provide a unified framework for learning hybrid dynamical systems from samples, contributing to advancements in fields such as cyber-physical systems, switching systems, robotics, and power systems.
Recommended Citation
Kaya, Ezgi, "An Exploration of Learning Hybrid Dynamical Models from Observations" (2024). Master's Theses. 5512.
DOI: https://doi.org/10.31979/etd.vswf-qfvk
https://scholarworks.sjsu.edu/etd_theses/5512
