Papers:
2024. G. Burgess, D.I. Singham, L. Rhodes-Leader. Time-Varying Capacity Planning for Designing Large-Scale Homeless Care Systems. Under Review. [PDF]
PhD project: Multi-fidelity modelling and optimisation with applications in long-term public sector capacity planning
Long-term capacity planning is a challenge for organisations worldwide, especially for public sector bodies who must meet basic public needs with tight financial budgets. Examples of such problems arise in housing, healthcare and prisons. In housing, decision makers must split their resources between emergency shelter and permanent social housing to alleviate homelessness. In healthcare, planners must balance the need for critical and non-critical care services when designing new hospitals. In prisons, capacity must be sufficient to avoid needing to release prisoners early but not so high as to risk costly unused capacity.
Operational research methods offer helpful tools to support such public sector decision-making. Optimisation helps by looking for a feasible solution which performs best across a set of alternative feasible solutions. To do this, a model of the system’s performance is needed, and the quality of the model affects the quality of the subsequent optimisation results. The most accurate models of the flow of people through public sector services are often high-fidelity stochastic simulation models. In this case, one can only estimate the performance of a given solution and the subsequent optimisation falls in the realm of simulation optimisation (SO). Low-fidelity models such as analytical queueing models offer a computationally cheaper alternative to stochastic simulation. They also offer helpful alternative perspectives on the dynamics of a queueing system. The drawback is that these models are typically less accurate, given the necessary assumptions which must be made. If one only uses a low-fidelity deterministic model to evaluate system performance, optimisation falls in the realm of deterministic optimisation. Performing this deterministic optimisation can be a helpful first step in the decision-making process. However, there is more we can do with low-fidelity models.
Multi-fidelity simulation optimisation (MFSO) enables low-fidelity models to be used alongside high-fidelity stochastic simulation in SO which reduces the computational burden and therefore enables an optimal solution to be found more efficiently. There are several technical challenges associated with MFSO when the solution space is integer-ordered and when low-fidelity queueing models give cheap information about the structure of this solution space. These conditions are often encountered in long-term capacity planning problems. As such, novel MFSO methods which tackle these technical challenges will be the main topic of this PhD research. We use a long-term public sector capacity planning problem as an example application.
Applying MFSO methods to long-term public sector capacity planning problems poses further challenges. If service times are long (as in housing or in prisons), uncertainty in the input models can overshadow stochastic uncertainty. An extension of our MFSO research will be to appropriately incorporate this input uncertainty. Secondly, with long planning horizons, decision makers typically revisit and adjust their plans on a regular basis, in light of new information. This dynamic element to the decision-making process is rarely addressed in the SO literature and will motivate another extension to our MFSO research.