Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Computing | en_US |
| dc.contributor.advisor | Wang, Dan (COMP) | en_US |
| dc.creator | Deng, Yang | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13380 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Towards AI deployment of the machine learning-based forecasting model in smart buildings | en_US |
| dcterms.abstract | Buildings are a major energy consumers and carbon emitters in modern society. In the US, buildings account for over 40% of total energy usage. To better operate building systems, and consequently reduce energy usage and the associated carbon emissions, the forecasting model, particularly, the load forecasting model that predicts the energy consumption or cooling/heating demand, plays an important role in HVAC control optimization, thermal energy storage operation, energy distribution system planning, and reliability control, to name but a few. | en_US |
| dcterms.abstract | With the development of IoT and AI technologies, building automation systems (BAS) are transforming into information-supported decision-making systems. This provides ample opportunities to develop data-driven machine learning (ML)-based forecasting models for downstream building applications. Typically, value is generated as soon as these models are deployed and continuously used in the BASs. However, in the building automation community, machine learning endeavors predominantly focus on conceiving applications for static situations (for example, ML feature engineering or the neural network design). In this context, the management of the ML models’ deployment to preserve their effectiveness over time in dynamic environments is still in its infancy. | en_US |
| dcterms.abstract | This dissertation posits that two key elements should be carefully considered while deploying the ML-based forecasting model: i) the model evaluation, which determines whether the AI forecasting model is suitable for the target building; and ii) the data representing the building, which is used to train and test the model. Therefore, this thesis starts with systematically analyzing the metrics of ML model deployment from both building and ML perspectives. On that basis, we then generate synthetic data to support the ML model under different scenarios via data augmentation methods. Finally, an AI evaluation platform involving the aforementioned components is developed. | en_US |
| dcterms.abstract | The contributions of this research are as follows, and we use building load forecasting to evaluate the proposed approaches. (1) We propose BuildChecks, a behavior testing methodology to systematically evaluate building load forecasting ML models in usage. The challenge of developing such a methodology is to specify ”what to evaluate”, i.e., given a certain building load forecasting model, what tests we shall apply to this model. We categorize three model-types of the building load forecasting models and we propose three in-usage concerns. Our methodology specifies the tests, i.e., for each model-type, the in-usage concerns that should be tested. (2) We propose DAST, a decomposition-based data augmentation scheme for insufficient data distribution, i.e., a data shortage caused by insufficient time coverage, a situation that is common in new buildings. We analyze data from hundreds of buildings to understand whether and what components exist. This analysis gives us prior knowledge of the decomposable components. We then develop DAST with appropriately designed decomposition, augmentation, and combination schemes. (3) We propose CADA, a concept drift-aware automated data augmentation model for establishing the suitable updating set for the ML model in the online scenario. This new data augmentation model is based on an ensemble concept drift detection algorithm to detect the three different types of concept drifts in the load data stream, and an automated data augmentation submodel to learn the data transformation policy under the drifts. (4) We developed a web platform with visualization and back-end services that contains the above research works. This aims to enable non-technical building employees to use AI techniques to benefit their applications. | en_US |
| dcterms.extent | xvi, 115 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2024 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Buildings -- Energy conservation | en_US |
| dcterms.LCSH | Buildings -- Environmental aspects | en_US |
| dcterms.LCSH | Buildings -- Energy conservation -- Automation | en_US |
| dcterms.LCSH | Artificial intelligence | en_US |
| dcterms.LCSH | Machine learning | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13380