• Login
• Admin

Hey, James (2022) Investigating the use of Machine Learning for Resource Intensive Agent-Based Simulation and Optimisation of Domestic Energy Retrofit Adoption. PhD thesis, University of Nottingham.

Preview

PDF (Thesis with corrections following examiners report) (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader Available under Licence Creative Commons Attribution. Download (6MB) | Preview

Abstract

The existing domestic urban built environment contributes significantly to the environ- mental issues facing the international community. Whole house energy retrofits per- formed on this stock are a key tool for the mitigation of greenhouse gas emissions. As such, researchers have built a set of bottom-up retrofit adoption models to understand, model, and predict household energy use and retrofit decisions in this space. However, the existing methods are limited in their ability to estimate complex decisions for large housing stocks due to computational restraints of optimisation. While machine learning has been applied to the problem of retrofit optimisation for performance improvements, the methods still scale poorly when applied to bottom-up agent-based models due to the large number of heterogeneous problems to be solved. This thesis aims to advance this toolset by extending state of the art data science approaches to domestic retrofit decision modelling across urban housing stocks.

The major contribution focuses on a transparent method of obtaining rapid predictions for near-optimal energy retrofit solutions using deep neural network models. This process is referred to as surrogate optimisation due to the surrogate modelling techniques used to achieve it. The models are trained on a sample of near-optimal solutions generated using traditional surrogate energy models paired with optimisation techniques to obtain a data set of retrofit decisions for model training. This allows for rapid estimation of retrofit decisions based on both the physical characteristics of the dwelling and the social characteristics of the households that would not be computationally feasible using existing methods.

This process is initially limited to the single objective of net present value to model rational and self-interested agents. The process is then extended to a multi-objective problem by considering net carbon emissions savings. This allows manipulation of the objective function to expose the household emissions valuation, which represents the marginal financial value a household places on each ton of carbon. By training the surrogate optimiser with these values, it was possible to both generate Pareto Fronts and target a specific carbon value held by a household, a characteristic that is both measurable and simple to understand. An agent-based model was constructed using the survey data derived decision model with the household carbon value trained surrogate optimiser. This allowed the consideration of scenarios and policy at a scale and level of detail which, without this research, would have been unfeasible with the computational resources used. After demonstrating the surrogate optimisation technique we perform a novel analysis of Best-Worst Scaling survey data in an attempt to understand household decisions better, ultimately resulting in a retrofit decision trigger model based on the responses. The more realistic trigger model was used in conjunction with the surrogate optimiser, capturing more realistic investment decisions.

Item Type:	Thesis (University of Nottingham only) (PhD)
Supervisors:	Siebers, Peer-Olaf Ozcan, Ender Nathanail, Paul Robinson, Darren
Keywords:	energy use, energy use modelling, machine learning, neural networks
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA 75 Electronic computers. Computer science
Faculties/Schools:	UK Campuses > Faculty of Science > School of Computer Science
Item ID:	69527
Depositing User:	Hey, James
Date Deposited:	27 Feb 2024 08:26
Last Modified:	27 Feb 2024 08:26
URI:	https://eprints.nottingham.ac.uk/id/eprint/69527

Actions (Archive Staff Only)

Edit View