Files in this item



application/pdfAMASYALI-DISSERTATION-2019.pdf (5MB)Restricted Access
(no description provided)PDF


Title:Machine learning for human-centered and value-sensitive building energy efficiency
Author(s):Amasyali, Kadir
Director of Research:El-Gohary, Nora
Doctoral Committee Chair(s):El-Gohary, Nora
Doctoral Committee Member(s):El-Rayes, Khaled; Liu, Liang Y.; Golparvar-Fard, Mani; Anumba, Chimay J
Department / Program:Civil & Environmental Eng
Discipline:Civil Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Building energy efficiency
Machine learning
Occupant comfort
Building energy prediction
Abstract:Enhancing building energy efficiency is one of the best strategies to reduce energy consumption and associated CO2 emissions. Recent studies emphasized the importance of occupant behavior as a key means of enhancing building energy efficiency. However, it is also critical that while we strive to enhance the energy efficiency of buildings through improving occupant behavior, we still pay enough attention to occupant comfort and satisfaction. Towards this goal, this research proposes a data-driven machine-learning-based approach to behavioral building energy efficiency, which could help better understand and predict the impact of occupant behavior on building energy consumption and occupant comfort; and help optimize occupant behavior for both energy saving and occupant comfort. Three types of models were developed and tested – simulation-data-driven, real-data-driven, and hybrid. Accordingly, the research included five primary research tasks. First, the importance levels of energy-related human values (e.g., thermal comfort) to building occupants and their current satisfaction levels with these values were identified, in order to better understand the factors that are associated with higher/lower importance and/or satisfaction levels and identify the potential factors that could help predict occupant comfort. Second, a data sensing and occupant feedback collection plan was developed, in order to capture and monitor the indoor environmental conditions, energy consumption, energy-related occupant behavior, and occupant comfort in real buildings. Third, a set of buildings were simulated, in order to model the energy consumption of different buildings in different contexts – in terms of occupant behavior, building sizes, weather conditions, etc.; and a simulation-data-driven occupant-behavior-sensitive machine learning-based model, which learns from simulation data, was developed for predicting hourly cooling energy consumption. Fourth, a set of real-data-driven occupant-behavior-sensitive machine learning-based models, which learn from real data (data collected from real buildings and real occupants), were developed for predicting hourly cooling and lighting energy consumption and thermal and visual occupant comfort; and a genetic algorithm-based optimization model for determining the optimal occupant behavior that can simultaneously reduce energy consumption and improve occupant comfort was developed. Compared to the simulation-data-driven approach, the real-data-driven approach aims to better capture and model the real-life behavior and comfort of occupants and the real-life energy-consumption patterns of buildings. Although successful in this regard, the resulting models may not generalize well outside of their training range. Fifth, a hybrid, occupant-behavior-sensitive machine learning-based model, which learns from both simulation data and real data, was developed for predicting hourly cooling and lighting energy consumption. The hybrid approach aims to overcome the limitations of both simulation-data-driven and real-data-driven approaches – especially the limited ability to capture occupant behavior and real-life consumption patterns in simulation-data-driven approaches and the limited generalizability of real-data-driven approaches to different cases – by learning from both types of data simultaneously. The experimental results show the potential of the proposed approach. The energy consumption prediction models achieved high prediction performance, and the thermal and visual comfort models were able to accurately represent the individual and group comfort levels. The optimization results showed potential behavioral energy savings in the range of 11% and 22%, with significant improvement in occupant comfort.
Issue Date:2019-04-19
Rights Information:Copyright 2019 Kadir Amasyali
Date Available in IDEALS:2019-08-23
Date Deposited:2019-05

This item appears in the following Collection(s)

Item Statistics