| Author: | Zhao, Lige |
| Title: | Input analysis for developing data-driven building energy prediction models |
| Advisors: | Xiao, Linda (BSE) |
| Degree: | M.Eng. |
| Year: | 2020 |
| Subject: | Buildings -- Energy consumption Buildings -- Energy consumption -- Data processing Buildings -- Energy conservation Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Building Services Engineering |
| Pages: | viii, 97 pages : color illustrations |
| Language: | English |
| Abstract: | With global urbanization, the rapid increase of building areas is accompanied by huge building energy consumption, which accounts for 36% of the total global energy use. Concerning of this great amount of energy consumption and the depletion of natural resources, recently researchers are making efforts to achieve a more sustainable way to operate and manage the building energy use. To get a better understanding of building energy consumption, prediction of building energy consumption the first step because the building prediction has been widespread used in developing new strategies for improving building performance such as fault detection and diagnosis (FDD), operation optimization. The conventional method of prediction is using the physical model, which named white box model and requires large amounts of inputs and the researchers should be equipped with building physics expertise which is very inconvenient and will cause a high cost. Development of intelligent computational efficient methods has contributed to the invention and application of data-driven methods, named black box and partial data driven methods, named grey box, which need to be combined with part of priori knowledge. ,more and more researchers built statistic models to achieve precise and easy-to-use prediction, for the purpose of obtaining a more convenient way of FDD and building performance optimization. And various research has studied on different aspects of building system big data, including the cross-sectional relationships and time series knowledge discovery. However, researchers are often working on data mining methods to develop more applicable algorithms or search for more cases to be study, but seldom focus on the primary question, i.e. selecting the right inputs for the models, which is even more important than the model itself. Considering all the facts above, this research mainly studied on the time series building energy prediction and developed a better way for input variable selection. In this study, BAS operation data of Hong Kong International Commercial Center building are combined, and seven popular feature selection algorithms are studied and analyzed by means of python programming. In the research process, the data of ICC building was preprocessed first, and the influence of historical cooling load data on energy consumption data at the present moment was explored by ACF and PACF methods, and the conclusion was drawn that the influence time range was about 15 hours. Therefore, the data 15 hours before the time is now treated as part of the candidate input variable pool. Then, the candidate input variable pool is input into seven algorithms for calculation, and the top 20 variables that have the greatest impact on the output result tot_c_load are selected by using each different method. Finally, these variables were input into the ANN model for prediction, and the results were used to evaluate the effect of the seven feature selection algorithms. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 5347.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 3.41 MB | Adobe PDF | View/Open |
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/200/10888