Gaussian Process Regression for Data-Driven Model Predictive Control

Gaussian Process Regression for Data-Driven Model Predictive Control

Seco Rodrigues, Nuno António (TU Delft Mechanical, Maritime and Materials Engineering)

van Wingerden, J.W. (mentor)

Delft University of Technology

Mechanical Engineering | Systems and Control

2021-06-06

Systems and Control deals with modelling and control design of many different types of systems with different behaviour and characteristics. Often times it is challenging to build mathematical models to describe the system dynamics due to limited knowledge of the underlying system, complex nonlinearities, or the existence of unknown exogenous phenomena impacting the input/output response. For this reason, it's important to develop methods of System Identification to learn these dynamics using only batches of input/output data, as well as control strategies capable of handling such models. The necessity for flexibility in the model and ability to deal with nonlinear mappings has led to the interest of applying machine learning methods to Identification and control design. This thesis focuses on using Gaussian Processes as a means to learn a model from batch training data and on developing a control strategy capable of handling such a model. The goal is to propose a Data-driven algorithm for closed-loop control and to explore the properties of a Gaussian Process that make it a candidate method for control applications.

Gaussian process
Regression
system identification
Machine Learning
Control
Online

http://resolver.tudelft.nl/uuid:67f90e43-9f4d-46dd-b4c9-6e002cb00ea3

Student theses

master thesis

© 2021 Nuno António Seco Rodrigues