Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis

Sze Yong, Necmiye Ozay

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we consider adaptive decision-making problems for stochastic discrete state estimation with a given budget of sensing actions/measurements that yield noisy and/or faulty partial observations. This problem is an extension of Bayesian active diagnosis, which is known to be NP-hard, to the setting when the sensor measurements are vector-valued and may be affected by persistent sensor faults and/or non-persistent noise. In particular, we identify meaningful reward functions for this problem that are adaptive monotone and weakly adaptive submodular; thus an adaptive greedy algorithm (with no need for proxy reward functions nor new algorithms) has guaranteed near-optimal performance. Finally, we apply our approach to discrete state estimation via active sensing of an electrical power system with sensor faults (persistent noise) and sensor noise (stochastic/non-persistent noise).

Original languageEnglish (US)
Title of host publication2018 Annual American Control Conference, ACC 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages313-320
Number of pages8
Volume2018-June
ISBN (Print)9781538654286
DOIs
StatePublished - Aug 9 2018
Event2018 Annual American Control Conference, ACC 2018 - Milwauke, United States
Duration: Jun 27 2018Jun 29 2018

Other

Other2018 Annual American Control Conference, ACC 2018
CountryUnited States
CityMilwauke
Period6/27/186/29/18

Fingerprint

State estimation
Sensors
Adaptive algorithms
Decision making

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Cite this

Yong, S., & Ozay, N. (2018). Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis. In 2018 Annual American Control Conference, ACC 2018 (Vol. 2018-June, pp. 313-320). [8431639] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.23919/ACC.2018.8431639

Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis. / Yong, Sze; Ozay, Necmiye.

2018 Annual American Control Conference, ACC 2018. Vol. 2018-June Institute of Electrical and Electronics Engineers Inc., 2018. p. 313-320 8431639.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Yong, S & Ozay, N 2018, Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis. in 2018 Annual American Control Conference, ACC 2018. vol. 2018-June, 8431639, Institute of Electrical and Electronics Engineers Inc., pp. 313-320, 2018 Annual American Control Conference, ACC 2018, Milwauke, United States, 6/27/18. https://doi.org/10.23919/ACC.2018.8431639
Yong S, Ozay N. Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis. In 2018 Annual American Control Conference, ACC 2018. Vol. 2018-June. Institute of Electrical and Electronics Engineers Inc. 2018. p. 313-320. 8431639 https://doi.org/10.23919/ACC.2018.8431639
Yong, Sze ; Ozay, Necmiye. / Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis. 2018 Annual American Control Conference, ACC 2018. Vol. 2018-June Institute of Electrical and Electronics Engineers Inc., 2018. pp. 313-320
@inproceedings{a6d2c0d666d0461c94f409821e8c32c4,
title = "Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis",
abstract = "In this paper, we consider adaptive decision-making problems for stochastic discrete state estimation with a given budget of sensing actions/measurements that yield noisy and/or faulty partial observations. This problem is an extension of Bayesian active diagnosis, which is known to be NP-hard, to the setting when the sensor measurements are vector-valued and may be affected by persistent sensor faults and/or non-persistent noise. In particular, we identify meaningful reward functions for this problem that are adaptive monotone and weakly adaptive submodular; thus an adaptive greedy algorithm (with no need for proxy reward functions nor new algorithms) has guaranteed near-optimal performance. Finally, we apply our approach to discrete state estimation via active sensing of an electrical power system with sensor faults (persistent noise) and sensor noise (stochastic/non-persistent noise).",
author = "Sze Yong and Necmiye Ozay",
year = "2018",
month = "8",
day = "9",
doi = "10.23919/ACC.2018.8431639",
language = "English (US)",
isbn = "9781538654286",
volume = "2018-June",
pages = "313--320",
booktitle = "2018 Annual American Control Conference, ACC 2018",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - GEN

T1 - Discrete State Estimation with Persistent Sensor Faults and Non-Persistent Noise via Noisy Bayesian Active Diagnosis

AU - Yong, Sze

AU - Ozay, Necmiye

PY - 2018/8/9

Y1 - 2018/8/9

N2 - In this paper, we consider adaptive decision-making problems for stochastic discrete state estimation with a given budget of sensing actions/measurements that yield noisy and/or faulty partial observations. This problem is an extension of Bayesian active diagnosis, which is known to be NP-hard, to the setting when the sensor measurements are vector-valued and may be affected by persistent sensor faults and/or non-persistent noise. In particular, we identify meaningful reward functions for this problem that are adaptive monotone and weakly adaptive submodular; thus an adaptive greedy algorithm (with no need for proxy reward functions nor new algorithms) has guaranteed near-optimal performance. Finally, we apply our approach to discrete state estimation via active sensing of an electrical power system with sensor faults (persistent noise) and sensor noise (stochastic/non-persistent noise).

AB - In this paper, we consider adaptive decision-making problems for stochastic discrete state estimation with a given budget of sensing actions/measurements that yield noisy and/or faulty partial observations. This problem is an extension of Bayesian active diagnosis, which is known to be NP-hard, to the setting when the sensor measurements are vector-valued and may be affected by persistent sensor faults and/or non-persistent noise. In particular, we identify meaningful reward functions for this problem that are adaptive monotone and weakly adaptive submodular; thus an adaptive greedy algorithm (with no need for proxy reward functions nor new algorithms) has guaranteed near-optimal performance. Finally, we apply our approach to discrete state estimation via active sensing of an electrical power system with sensor faults (persistent noise) and sensor noise (stochastic/non-persistent noise).

UR - http://www.scopus.com/inward/record.url?scp=85052573514&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85052573514&partnerID=8YFLogxK

U2 - 10.23919/ACC.2018.8431639

DO - 10.23919/ACC.2018.8431639

M3 - Conference contribution

AN - SCOPUS:85052573514

SN - 9781538654286

VL - 2018-June

SP - 313

EP - 320

BT - 2018 Annual American Control Conference, ACC 2018

PB - Institute of Electrical and Electronics Engineers Inc.

ER -