Emulation of Stochastic Computer Models with an Application to Building Design
Lead Research Organisation:
UNIVERSITY OF EXETER
Department Name: Engineering Computer Science and Maths
Abstract
Climate change over the next few decades will affect the way we design buildings. In particular improved insulation and rising temperatures could mean that overheating becomes a serious problem, in both domestic and public buildings (particularly in hospitals). In this PhD we will study the effect of climate change on the thermal properties of buildings. We will use mathematical models of buildings driven by stochastic weather files. These allow us to model the variability in weather. Running the full numerical model of a building is computationally expensive so we build an emulator for this model. An emulator is a fast statistical approximation to the full expensive numerical code. Because we are using stochastic weather fields to drive the model we will need to derive new stochastic emulators that preserve the uncertainty coming from the weather files. Once we have built the emulator we will explore the effect of different future climates and building designs on thermal performance (and overheating). We will investigate the possibility of optimal design for the thermal performance of buildings which will last for the next 50-100 years.
Publications
Baker E
(2021)
Future Proofing a Building Design using History Matching Inspired Level-set Techniques
in Journal of the Royal Statistical Society Series C: Applied Statistics
Baker E
(2020)
Predicting the Output From a Stochastic Computer Model When a Deterministic Approximation is Available
in Journal of Computational and Graphical Statistics
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
EP/N509656/1 | 30/09/2016 | 29/09/2021 | |||
1917423 | Studentship | EP/N509656/1 | 30/09/2017 | 07/09/2021 | Evan Baker |
Description | Emulators are statistical surrogate models used to act in lieu of full, computationally expensive, computer models. During this PhD, new methodology for building emulators was developed, wherein the outputs from a deterministic (i.e. non-noisy) computer model can be linked to the outputs from a stochastic (noisy) computer model. Additionally, a framework for optimising the insulatory properties of a building was developed and showcased (using a complex energy usage model called EnergyPlus, coupled with emulators). Two journal articles were published as a result of this work. Additionally, a collaborative project with the American institution SAMSI resulted in a publication explaining how one can build and use emulators for stochastic computer models. |
Exploitation Route | Academically, the publications directly obtained from the PhD could be taken further. Linking deterministic and stochastic models together to obtain improved results is an interesting avenue for future research, both methodologically and implementing in practice. The review done with aid from SAMSI is a valuable tool for new researchers and practioners who have access to a stochastic computer model and need assistance in performing analysis. Additionally, the framework developed in the second publication could see use in industry, and seems off-the-shelf capable. |
Sectors | Energy Other |
Description | Collaboration with SAMSI |
Organisation | Statistical and Applied Mathematical Sciences Institute |
Country | United States |
Sector | Academic/University |
PI Contribution | Together, with other academics that attended a SAMSI programme, we wrote and published a review paper |
Collaborator Contribution | Collaborators helped write and provide content for the review paper |
Impact | Review paper on statistical emulation for stochastic computer models was written and published: https://projecteuclid.org/journals/statistical-science/volume-37/issue-1/Analyzing-Stochastic-Computer-Models-A-Review-with-Opportunities/10.1214/21-STS822.short. In a way, this is cross disciplinary, including statistics, mathematical modelling, oceanography, ecology, epidimieology, and more. |
Start Year | 2019 |