Abstract
The design stage of a building plays a pivotal role in influencing its life cycle and overall performance. Accurate predictions of a building's performance are crucial for informed decision-making, particularly in terms of energy performance, given the escalating global awareness of climate change and the imperative to enhance energy efficiency in buildings. However, a well-documented energy performance gap persists between actual and predicted energy consumption, primarily attributed to the unpredictable nature of occupant behavior.Existing methodologies for predicting and simulating occupant behavior in buildings frequently neglect or exclusively concentrate on particular behaviors, resulting in uncertainties in energy performance predictions. Machine learning approaches have exhibited increased accuracy in predicting occupant energy behavior, yet the majority of extant studies focus on specific behavior types rather than investigating the interactions among all contributing factors. This dissertation delves into the building energy performance gap, with a particular emphasis on the influence of occupants on energy performance. A comprehensive literature review scrutinizes machine learning models employed for predicting occupants' behavior in buildings and assesses their performance. The review uncovers knowledge gaps, as most studies are case-specific and lack a consolidated database to examine diverse behaviors across various building types.
An ensemble model integrating occupant behavior parameters is devised to enhance the accuracy of energy performance predictions in residential buildings. Multiple algorithms are examined, with the selection of algorithms contingent upon evaluation metrics. The ensemble model is validated through a case study that compares actual energy consumption with the predictions of the ensemble model and an EnergyPlus simulation that takes occupant behavior factors into account.
The findings demonstrate that the ensemble model provides considerably more accurate predictions of actual energy consumption compared to the EnergyPlus simulation. This dissertation also addresses the research limitations, including the reusability of the model and the requirement for additional datasets to bolster confidence in the model's applicability across diverse building types and occupant behavior patterns.
In summary, this dissertation presents an ensemble model that endeavors to bridge the gap between actual and predicted energy usage in residential buildings by incorporating occupant behavior parameters, leading to more precise energy performance predictions and promoting superior energy management strategies.
| Date of Award | 21 Nov 2023 |
|---|---|
| Original language | English |
| Supervisor | Patricia Tzortzopoulos (Main Supervisor) & Elham Delzendeh (Co-Supervisor) |