Multi-objective optimization for sensor placement: An integrated combinatorial approach with reduced order model and Gaussian process

•We devise a multi-objective optimization (MOO) for sensor placement.•Our MOO integrates reduced order model and lazy greedy combinatorial approach.•We develop branch and bound exact method to validate the Pareto frontier.•We validate our method by a temperature sensor placement example. We develop...

Full description

Saved in:
Bibliographic Details
Published in:Measurement : journal of the International Measurement Confederation Vol. 187; p. 110370
Main Authors: Xu, Zhaoyi, Guo, Yanjie, Homer Saleh, Joseph
Format: Journal Article
Language:English
Published: London Elsevier Ltd 01.01.2022
Elsevier Science Ltd
Subjects:
Algorithms
Branch and bound
Combinatorial analysis
Configuration management
Gaussian process
Lazy greedy algorithm
Multi-objective combinatorial optimization
Multiple objective analysis
Normal distribution
Optimization
Pareto optimization
Pareto optimum
Placement
Proper Orthogonal Decomposition
Reduced order models
Sensor placement
Sensors
State estimation
Sensor placement
Proper orthogonal decomposition
Lazy greedy algorithm
Branch and bound
Multi-objective combinatorial optimization
ISSN:0263-2241, 1873-412X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•We devise a multi-objective optimization (MOO) for sensor placement.•Our MOO integrates reduced order model and lazy greedy combinatorial approach.•We develop branch and bound exact method to validate the Pareto frontier.•We validate our method by a temperature sensor placement example. We develop a novel sensor placement method that maximizes monitoring performance while minimizing deployment cost. Our method integrates a reduced order model and multi-objective combinatorial optimization. We first decompose the spatio-temporal state field to be monitored by proper orthogonal decomposition (POD), and we use the Gaussian Process to model the uncertainty in each POD mode. Next, we develop a lazy greedy (LG)-∊-constraint optimization to derive the Pareto-optimal sensor configurations. We further design a branch and bound algorithm to calculate the global optimum and validate the correctness of select configurations on the LG-derived Pareto frontier. We evaluate and benchmark our algorithm in computational experiments based on the temperature dataset of the Berkeley Intel Lab. The computational results demonstrate that our algorithm places sensors at locations of large magnitude in the POD modes, and that our method achieves better state estimation accuracy and smaller reconstruction errors compared with alternative methods.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0263-2241
1873-412X
DOI:10.1016/j.measurement.2021.110370