Using Deep Learning to Control of Complex Systems

Please enable javascript in your browser.

Aminorroaya, Saba | 2022

117 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 55339 (04)
University: Sharif University of Technology
Department: Physics
Advisor(s): Rahimi Tabar, Mohammad Reza
Abstract:
A complex system consists of a large number of subsystems that interact with each other and with the environment. These systems have collective behaviors that may are desired and undesired. Learning, intelligence and epilepsy are examples of desirable and undesirable collective behaviors. Control of these systems arises when they are out of the desired state or one wants to avoid approaching the system to its undesired state. For control of complex systems, we need external functions that apply to specific subsystems. These functions can be obtained from the numerical solution of Hamilton-Jacobi-Bellman equation. The Hamilton-Jacobi-Bellman equation is nonlinear and must be solved at very high dimensions due to the high dimensions of complex systems. In this thesis, by reviewing the deep learning methods in solving the HamiltonJacobi-Bellman equation, in one example, the deep learning method will be compared with the numerical method.
Keywords:
Complex System ; Deep Learning ; Hamilton-Jacobi-Bellman Equation ; Complex Systems Control

No TOC