Model free real-time optimization for vapor compression systems

Keating, Bryan David

Permalink

https://hdl.handle.net/2142/97233 Copy

Description

Title

Model free real-time optimization for vapor compression systems

Author(s)

Keating, Bryan David

Issue Date

2016-11-23

Director of Research (if dissertation) or Advisor (if thesis)

Alleyne, Andrew G.

Department of Study

Mechanical Sci & Engineering

Discipline

Mechanical Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Optimization
• Extremum seeking
• Vapor compression systems

Abstract

A vapor compression system's optimal input settings vary according to changes in environmental conditions. Tracking these optimal input trajectories can be challenging when insufficient information for a reliable system model is available. An alternative set of optimization approaches use system measurements. This thesis focuses on one such approach, extremum seeking control, which uses performance index measurements to determine optimal system settings. Forgoing system model knowledge and relying exclusively on data allows an optimization approach to function well on many different plants. However, this added adaptivity comes at a performance cost. Using prior system model knowledge can be helpful for ensuring that a controller design works from the start of operation and inputs can be changed as soon as information about environmental conditions is updated. By contrast, data based methods may require the control designer to spend a time generating data in order to obtain enough information about the system to make good decisions online. A central theme of this work is addressing the trade off between using prior system model knowledge and ensuring sufficient adaptability of the extremum seeking optimization approach. Two main factors in the extremum seeking design are considered: the choice of extremum seeking control law and the choice of extremum seeking control input. Extremum seeking control laws come from the field of mathematical optimization; this thesis considers the pros and cons of choosing between gradient descent and Newton descent. Both simulations and experimental results show that while Newton descent extremum seeking is less reliant on model knowledge, but slower to find optimal inputs than gradient descent extremum seeking. Because of extremum seeking's adaptability to different plants, many different inputs can be chosen for implementation. However, using an approach known as self-optimizing control, knowledge about the plant's behavior can help choose set points with optimal values that are insensitive to changes in environmental conditions. Finding these special inputs turns the input tracking problem into a regulation problem. Both simulation and experimental results confirm that combining self-optimizing control and extremum seeking control can help improve tracking even as environmental conditions change.

Graduation Semester

2017-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/97233

Copyright and License Information

Copyright 2017 Bryan Keating

Owning Collections