Kinetic models and intrinsic timescales

Simulation comparison for a 2nd order queueing model

Hans Armbruster, Matthew Wienke

Research output: Contribution to journalArticle

Abstract

Kinetic models of stochastic production ows can be expanded into deterministic moment equations and thus approximated with appropriate closures. A second order model for the product density and the product speed has previously been proposed. A systematic analysis comparing simulations of the partial differential equations (PDE) with discrete event simulations (DES) is performed. Specifically, factory production is modeled as an M/M/1 queue where the arrival process is a non-homogeneous Poisson process. Three fundamental scenarios for such a time dependent in ux are studied: An instant step up/step down of the arrival rate, an exponential step up/step down and periodic variation of the average arrival rate. It is shown that the second order model in general yields significant improvements over the first order model. Adding diffusion into the PDE further improves the agreement in particular for queues with low utilization. The analysis also points to fundamental open issues regarding kinetic models of time dependent agent based simulations. Memory effects and the possibility of resonance in deterministic models are caused by intrinsic timescales of the PDE that are not present in the original stochastic processes.

Original languageEnglish (US)
Pages (from-to)177-193
Number of pages17
JournalKinetic and Related Models
Volume12
Issue number1
DOIs
StatePublished - Feb 1 2019

Fingerprint

Queueing Model
Kinetic Model
Second-order Model
Time Scales
Partial differential equation
Kinetics
M/M/1 Queue
Partial differential equations
Non-homogeneous Poisson Process
Moment Equations
Simulation
Memory Effect
Agent-based Simulation
Deterministic Model
Simulation Analysis
Discrete Event Simulation
Instant
Queue
Stochastic Processes
Closure

Keywords

  • 2nd order models
  • Kinetic models
  • Production systems
  • Simulations
  • Transients

ASJC Scopus subject areas

  • Numerical Analysis
  • Modeling and Simulation

Cite this

Kinetic models and intrinsic timescales : Simulation comparison for a 2nd order queueing model. / Armbruster, Hans; Wienke, Matthew.

In: Kinetic and Related Models, Vol. 12, No. 1, 01.02.2019, p. 177-193.

Research output: Contribution to journalArticle

@article{05f2c4e7a3084906af32eda3264c8296,
title = "Kinetic models and intrinsic timescales: Simulation comparison for a 2nd order queueing model",
abstract = "Kinetic models of stochastic production ows can be expanded into deterministic moment equations and thus approximated with appropriate closures. A second order model for the product density and the product speed has previously been proposed. A systematic analysis comparing simulations of the partial differential equations (PDE) with discrete event simulations (DES) is performed. Specifically, factory production is modeled as an M/M/1 queue where the arrival process is a non-homogeneous Poisson process. Three fundamental scenarios for such a time dependent in ux are studied: An instant step up/step down of the arrival rate, an exponential step up/step down and periodic variation of the average arrival rate. It is shown that the second order model in general yields significant improvements over the first order model. Adding diffusion into the PDE further improves the agreement in particular for queues with low utilization. The analysis also points to fundamental open issues regarding kinetic models of time dependent agent based simulations. Memory effects and the possibility of resonance in deterministic models are caused by intrinsic timescales of the PDE that are not present in the original stochastic processes.",
keywords = "2nd order models, Kinetic models, Production systems, Simulations, Transients",
author = "Hans Armbruster and Matthew Wienke",
year = "2019",
month = "2",
day = "1",
doi = "10.3934/krm.2019008",
language = "English (US)",
volume = "12",
pages = "177--193",
journal = "Kinetic and Related Models",
issn = "1937-5093",
publisher = "American Institute of Mathematical Sciences",
number = "1",

}

TY - JOUR

T1 - Kinetic models and intrinsic timescales

T2 - Simulation comparison for a 2nd order queueing model

AU - Armbruster, Hans

AU - Wienke, Matthew

PY - 2019/2/1

Y1 - 2019/2/1

N2 - Kinetic models of stochastic production ows can be expanded into deterministic moment equations and thus approximated with appropriate closures. A second order model for the product density and the product speed has previously been proposed. A systematic analysis comparing simulations of the partial differential equations (PDE) with discrete event simulations (DES) is performed. Specifically, factory production is modeled as an M/M/1 queue where the arrival process is a non-homogeneous Poisson process. Three fundamental scenarios for such a time dependent in ux are studied: An instant step up/step down of the arrival rate, an exponential step up/step down and periodic variation of the average arrival rate. It is shown that the second order model in general yields significant improvements over the first order model. Adding diffusion into the PDE further improves the agreement in particular for queues with low utilization. The analysis also points to fundamental open issues regarding kinetic models of time dependent agent based simulations. Memory effects and the possibility of resonance in deterministic models are caused by intrinsic timescales of the PDE that are not present in the original stochastic processes.

AB - Kinetic models of stochastic production ows can be expanded into deterministic moment equations and thus approximated with appropriate closures. A second order model for the product density and the product speed has previously been proposed. A systematic analysis comparing simulations of the partial differential equations (PDE) with discrete event simulations (DES) is performed. Specifically, factory production is modeled as an M/M/1 queue where the arrival process is a non-homogeneous Poisson process. Three fundamental scenarios for such a time dependent in ux are studied: An instant step up/step down of the arrival rate, an exponential step up/step down and periodic variation of the average arrival rate. It is shown that the second order model in general yields significant improvements over the first order model. Adding diffusion into the PDE further improves the agreement in particular for queues with low utilization. The analysis also points to fundamental open issues regarding kinetic models of time dependent agent based simulations. Memory effects and the possibility of resonance in deterministic models are caused by intrinsic timescales of the PDE that are not present in the original stochastic processes.

KW - 2nd order models

KW - Kinetic models

KW - Production systems

KW - Simulations

KW - Transients

UR - http://www.scopus.com/inward/record.url?scp=85061896155&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85061896155&partnerID=8YFLogxK

U2 - 10.3934/krm.2019008

DO - 10.3934/krm.2019008

M3 - Article

VL - 12

SP - 177

EP - 193

JO - Kinetic and Related Models

JF - Kinetic and Related Models

SN - 1937-5093

IS - 1

ER -