東華大學圖書館 |

Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization./
Author:	Zhang, Ling.
Published:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
Description:	175 p.
Notes:	Source: Dissertations Abstracts International, Volume: 82-02, Section: B.
Contained By:	Dissertations Abstracts International82-02B.
Subject:	Industrial engineering. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28075129
ISBN:	9798662445734

Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization.
Zhang, Ling.

Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 175 p.

Source: Dissertations Abstracts International, Volume: 82-02, Section: B.

Thesis (Ph.D.)--North Carolina State University, 2020.

This item must not be sold to any third party vendors.

This thesis studies optimal control problems for service systems, such as manufacturing centers, healthcare facilities, and vehicle-sharing networks. It is commonly observed that exogenous demands arriving at these systems tend to vary with time in a finite horizon. Moreover, service providers often fail to timely and flexibly adjust service capacity in response to meet time-varying demand. Direct consequences of such inefficiency include undesired degradation of service quality (e.g., excessively long waiting queue) and inevitable wastage of service resources (e.g., idle servers). This thesis is motivated by such operational challenges created mainly by time-vary demands and aims to provide design and operational guidance for decision-makers.These service systems are first modeled as stochastic queueing networks. Next, optimal control problems are developed for the original stochastic models. However, due to factors such as time-varying demand, network structure, and other realistic features (e.g., customer abandonment, non-exponential patience and service times, capacity changing costs), these optimal control problems are not amenable to exact mathematical analysis. Therefore, relaxed optimal control problems based on continuous-time deterministic fluid approximations are studied instead. Pontryagin's Maximum Principle is used to solve fluid-based optimal control problems analytically. When the aforementioned realistic features forbid the direct application of the maximum principle, efficient heuristics are designed to obtain numerical solutions, and insights as well. Fluid optimal solutions provide us with structural results to guide the design of asymptotically fluid optimal control policies for the original stochastic systems.As demand rates and service rates increase to infinity, this thesis shows that the objective function of a stochastic system applied with the asymptotically fluid optimal control policy converges to that of its fluid counterpart applied with the fluid optimal solution. Furthermore, the fluid objective function applied with the optimal solution also serves as an asymptotic lower bound to the original stochastic problem applied with any admissible control policies. Simulation comparisons between the asymptotically fluid optimal policies and myopics policies also validate these analytical results.This thesis consists of three pieces of work, each presented in one chapter. The first chapter (Chapter 2) studies an optimal production control problem in a double-ended queue with a time-varying demand arrival rate and production rate. It is motivated by systems where a pair of demand and supply leaves the system together as soon as one unit of demand is matched with one unit of supply. Examples of such systems include pharmaceutical manufacturers with continuous production, make-to-order systems, online auction platforms, etc. Demands and supplies are both impatient with exponential patience times. More importantly, the controllable production process is costly to be flexible. The objective is to construct optimal production policies to minimize total cost, which includes production cost, inventory holding cost, wastage cost, backorder cost, demand abandonment cost, and production changing cost. A fluid-based optimal production rate control problem is developed and is shown to be the lower bound for the corresponding stochastic control problem. We also prove that the asymptotic optimality of fluid optimal control. Simulations are also conducted to validate these results. In order to obtain an implementable solution, discretization is applied, and the resulting discrete-time problem is a linear programming problem. To validate the asymptotic optimality of the discrete-time fluid-based solution, we compare its performance with the numerical solution to a dynamic programming formulation of the stochastic problem.The second chapter (Chapter 3) develops a closed queuing network to model a one-way car-sharing system. It is motivated by the booming sharing economy in the public transportation sector, especially by emerging companies such as Zipcar, Car2Go, and Uber. A time-varying demand rate characterizes the rush-hour phenomena exhibited in the customer arrival process. It is essential to consider this particular feature because the surge demand in rush hour tends to create unbalanced demand and supply, which will jeopardize the service quality and utilization of service resources in between rush hours. Additionally, the travel time distribution taking into account is non-Markovian. The network and non-Markovian structure of the problem suggests a treatment using the asymptotic fluid model. We provide a contraction-based algorithm to solve the fluid dynamics and validate the accuracy of the fluid model with simulation. Based on the fluid approximation of the closed queueing network, two fluid control problems are proposed. One is to maximize total system throughput, and the other one is to maximize total revenue. The decision variables in the fluid control problem are admission and pricing. Numerical heuristics to solve these two problems are proposed and simulation experiments are conducted to validate the effectiveness over myopic policies, such as the first-come-first-serve policy. More importantly, these results facilitate the exploration of insights on the time-heterogeneity of fluid optimal policies and highlight the importance of including information beyond the mean of travel times.The third chapter (Chapter 4) studies a tandem queueing network which models a twostation service system in an emergency department. In addition to time-varying demand, two realistic features are included in the model: customer splitting after treatment in the first queue and intermediate delay between two service stations. We proposed two optimization framework to control this tandem system based on different queueing characterization,i.e., many-server and single-server heavy-traffic fluid approximations. The target is to minimize waiting costs at both service stations by prescribing optimal staffing schedules. Heuristic control policies are derived from optimal solutions, such as a c-μ type of rule.We developed a discrete-event simulation that aims to characterize more realistic features to validate the effectiveness of different policies.

ISBN: 9798662445734Subjects--Topical Terms:

526216
Industrial engineering.
Subjects--Index Terms:

Time-varying service networks

Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization.
LDR:07584nmm a2200337 4500 001 2276028
005 20210416102016.5
008 220723s2020 ||||||||||||||||| ||eng d
020 $a 9798662445734
035 $a (MiAaPQ)AAI28075129
035 $a (MiAaPQ)NCState_Univ18402037998
035 $a AAI28075129
040 $a MiAaPQ $c MiAaPQ
100 1 $a Zhang, Ling. $3 1273486
245 1 0 $a Optimal Controls in Time-varying Service Networks: Asymptotic Effectiveness and Performance Optimization.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 175 p.
500 $a Source: Dissertations Abstracts International, Volume: 82-02, Section: B.
500 $a Advisor: Kang, Min;Argon, Nilay;Liu, Xin;Pang, Tao;Medhin, Negash;Fang, Shu-Cherng;Liu, Yunan.
502 $a Thesis (Ph.D.)--North Carolina State University, 2020.
506 $a This item must not be sold to any third party vendors.
520 $a This thesis studies optimal control problems for service systems, such as manufacturing centers, healthcare facilities, and vehicle-sharing networks. It is commonly observed that exogenous demands arriving at these systems tend to vary with time in a finite horizon. Moreover, service providers often fail to timely and flexibly adjust service capacity in response to meet time-varying demand. Direct consequences of such inefficiency include undesired degradation of service quality (e.g., excessively long waiting queue) and inevitable wastage of service resources (e.g., idle servers). This thesis is motivated by such operational challenges created mainly by time-vary demands and aims to provide design and operational guidance for decision-makers.These service systems are first modeled as stochastic queueing networks. Next, optimal control problems are developed for the original stochastic models. However, due to factors such as time-varying demand, network structure, and other realistic features (e.g., customer abandonment, non-exponential patience and service times, capacity changing costs), these optimal control problems are not amenable to exact mathematical analysis. Therefore, relaxed optimal control problems based on continuous-time deterministic fluid approximations are studied instead. Pontryagin's Maximum Principle is used to solve fluid-based optimal control problems analytically. When the aforementioned realistic features forbid the direct application of the maximum principle, efficient heuristics are designed to obtain numerical solutions, and insights as well. Fluid optimal solutions provide us with structural results to guide the design of asymptotically fluid optimal control policies for the original stochastic systems.As demand rates and service rates increase to infinity, this thesis shows that the objective function of a stochastic system applied with the asymptotically fluid optimal control policy converges to that of its fluid counterpart applied with the fluid optimal solution. Furthermore, the fluid objective function applied with the optimal solution also serves as an asymptotic lower bound to the original stochastic problem applied with any admissible control policies. Simulation comparisons between the asymptotically fluid optimal policies and myopics policies also validate these analytical results.This thesis consists of three pieces of work, each presented in one chapter. The first chapter (Chapter 2) studies an optimal production control problem in a double-ended queue with a time-varying demand arrival rate and production rate. It is motivated by systems where a pair of demand and supply leaves the system together as soon as one unit of demand is matched with one unit of supply. Examples of such systems include pharmaceutical manufacturers with continuous production, make-to-order systems, online auction platforms, etc. Demands and supplies are both impatient with exponential patience times. More importantly, the controllable production process is costly to be flexible. The objective is to construct optimal production policies to minimize total cost, which includes production cost, inventory holding cost, wastage cost, backorder cost, demand abandonment cost, and production changing cost. A fluid-based optimal production rate control problem is developed and is shown to be the lower bound for the corresponding stochastic control problem. We also prove that the asymptotic optimality of fluid optimal control. Simulations are also conducted to validate these results. In order to obtain an implementable solution, discretization is applied, and the resulting discrete-time problem is a linear programming problem. To validate the asymptotic optimality of the discrete-time fluid-based solution, we compare its performance with the numerical solution to a dynamic programming formulation of the stochastic problem.The second chapter (Chapter 3) develops a closed queuing network to model a one-way car-sharing system. It is motivated by the booming sharing economy in the public transportation sector, especially by emerging companies such as Zipcar, Car2Go, and Uber. A time-varying demand rate characterizes the rush-hour phenomena exhibited in the customer arrival process. It is essential to consider this particular feature because the surge demand in rush hour tends to create unbalanced demand and supply, which will jeopardize the service quality and utilization of service resources in between rush hours. Additionally, the travel time distribution taking into account is non-Markovian. The network and non-Markovian structure of the problem suggests a treatment using the asymptotic fluid model. We provide a contraction-based algorithm to solve the fluid dynamics and validate the accuracy of the fluid model with simulation. Based on the fluid approximation of the closed queueing network, two fluid control problems are proposed. One is to maximize total system throughput, and the other one is to maximize total revenue. The decision variables in the fluid control problem are admission and pricing. Numerical heuristics to solve these two problems are proposed and simulation experiments are conducted to validate the effectiveness over myopic policies, such as the first-come-first-serve policy. More importantly, these results facilitate the exploration of insights on the time-heterogeneity of fluid optimal policies and highlight the importance of including information beyond the mean of travel times.The third chapter (Chapter 4) studies a tandem queueing network which models a twostation service system in an emergency department. In addition to time-varying demand, two realistic features are included in the model: customer splitting after treatment in the first queue and intermediate delay between two service stations. We proposed two optimization framework to control this tandem system based on different queueing characterization,i.e., many-server and single-server heavy-traffic fluid approximations. The target is to minimize waiting costs at both service stations by prescribing optimal staffing schedules. Heuristic control policies are derived from optimal solutions, such as a c-μ type of rule.We developed a discrete-event simulation that aims to characterize more realistic features to validate the effectiveness of different policies.
590 $a School code: 0155.
650 4 $a Industrial engineering. $3 526216
653 $a Time-varying service networks
653 $a Optimal controls
653 $a Service systems
690 $a 0546
710 2 $a North Carolina State University. $3 1018772
773 0 $t Dissertations Abstracts International $g 82-02B.
790 $a 0155
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28075129