Dynamic Control of Multiclass Multi-Episode Service Systems with Waiting Time Dependent Transitions

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

University of Waterloo

Abstract

Motivated by rehabilitation services, this thesis studies a service system with multiple heterogeneous customer classes and servers in which customers may require multiple episodes of service to achieve a desired outcome. After each episode, the service outcome is observed, and customers who do not achieve the desired outcome transition to another class for additional service. Empirical evidence indicates that prolonged waiting can worsen service outcomes and increase future congestion. The model therefore allows service outcomes and class transitions to depend jointly on the service rate employed and the waiting time experienced. The system manager makes joint dynamic scheduling, service-rate, and customer-rejection decisions to minimize a long-run average expected cost comprising waiting, service, abandonment, rejection, and outcome-related transition costs, subject to firm throughput-time deadlines. Because the resulting stochastic control problem is analytically intractable, the thesis develops a Brownian approximation in a heavy-traffic regime. Under complete resource pooling, the Brownian control problem is reduced to a one-dimensional workload formulation. Its optimal solution is characterized as a two-sided barrier policy with a state-dependent drift rate and workload configuration. The workload solution is translated into a tractable control policy for the original service system. A comprehensive simulation study shows that the proposed policy substantially outperforms static and myopic benchmarks. The results demonstrate that explicitly accounting for waiting-time-dependent outcomes is essential: policies that ignore these effects can perform poorly because they optimize immediate costs without accounting for the effect of current waiting on future congestion. The analysis yields several structural insights. Scheduling and service-rate control affect congestion through distinct mechanisms and perform best when used jointly. Scheduling priorities balance immediate costs against downstream congestion effects, shifting toward classes whose waiting times most strongly amplify future congestion as the system becomes more congested. Service-rate control exhibits a regime-switching structure: under low congestion it emphasizes service outcomes, whereas under high congestion it emphasizes congestion mitigation. Finally, although abandonment is costly, it can partially offset the adverse congestion effects of waiting-time-dependent transitions and therefore acts as a partial substitute for dynamic service-rate control.

Description

Citation

Collections

Endorsement

Review

Supplemented By

Referenced By