Alleviating tuning sensitivity in Approximate Dynamic Programming


METADATA ONLY
Loading...

Date

2016

Publication Type

Conference Paper

ETH Bibliography

yes

Citations

Altmetric
METADATA ONLY

Data

Rights / License

Abstract

The simplicity of Approximate Dynamic Programming offers benefits for large-scale systems compared to other synthesis and control methodologies. A common technique to approximate the Dynamic Program, is through the solution of the corresponding Linear Program. The major drawback of this approach is that the online performance is very sensitive to the choice of tuning parameters, in particular the state relevance weighting parameter. Our work aims at alleviating this sensitivity. To achieve this, we propose to find a set of approximate Q-functions, each for a different choice of the tuning parameters, and then to use the pointwise maximum of the set of Q-functions for the online policy. The pointwise maximum promises to be better than using only one of individual Q-functions for the online policy. We demonstrate that this approach immunizes against tuning errors through a stylized portfolio optimization problem.

Publication status

published

Editor

Book title

2016 European Control Conference (ECC)

Journal / series

Volume

Pages / Article No.

1616 - 1622

Publisher

IEEE

Event

15th Annual European Control Conference (ECC 2016)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Organisational unit

03751 - Lygeros, John / Lygeros, John check_circle

Notes

Funding

Related publications and datasets