Multi-feature Time Series Forecasting for Household Energy Requirements
| dc.contributor.author | Mustapha, Akeem Amao | en |
| dc.date.accessioned | 2024-06-21T09:43:51Z | |
| dc.date.available | 2024-06-21T09:43:51Z | |
| dc.date.issued | 2024-06-21 | |
| dc.identifier.uri | https://repository.ihu.edu.gr//xmlui/handle/11544/30478 | |
| dc.rights | Default License | |
| dc.subject | Time series forecasting | en |
| dc.subject | Household energy requirements | en |
| dc.subject | Energy consumption | en |
| dc.subject | Machine learning model | en |
| dc.title | Multi-feature Time Series Forecasting for Household Energy Requirements | en |
| heal.abstract | To curb the effects of greenhouse gas emissions on our environment, there has been a call for the high-energy-consuming building sector to reduce energy consumption. Forecasting energy consumption in residential buildings is very important to achieve this. However, improving forecasting efficiency for short-term load forecasting has been the state-of-the-art focus on building energy forecasting due to several factors that affect energy consumption in buildings. As an explorative experiment, this study employed traditional machine learning models, which have proven effective for forecasting to predict very short-term energy consumption in ITI smart house in Thessaloniki. Three categories of machine learning models were selected: the linear, tree-based and boosting models. A comparative analysis of the performance of the chosen models based on different time steps (15-minute, 30-minute, 1-hour), engineered features and hyperparameter optimization is carried out to identify the key elements to obtain accurate predictions. The best-performing models are the boosting models, with LGBM outperforming all models in the experiment based on the performance evaluation metrics used. The performance of the linear models remained more or less the same throughout the experiments. Optimizing the hyperparameter to improve the prediction performance, XGBoost regression gave the best prediction based on MAE for a 1-hour time step. The overall best performance for the models was recorded for the 15-minute and 1-hour time steps. Analysing the importance of the features to the models showed the dependency of models on temperature as a weather factor and the predicting power of time-related predictors in their performance. Generally, the experiments revealed that appropriate data preprocessing is a significant process that determines the accuracy of prediction models and should be suited for the models in use. | en |
| heal.academicPublisher | IHU | en |
| heal.academicPublisherID | ihu | en_US |
| heal.access | free | en_US |
| heal.advisorName | Koukaras, Paraskevas | en |
| heal.committeeMemberName | Tjortjis, Christos | en |
| heal.creatorID.email | amustapha@ihu.edu.gr | |
| heal.dateAvailable | 2024-06-05 | |
| heal.language | en | en_US |
| heal.license | http://creativecommons.org/licenses/by-nc/4.0 | en_US |
| heal.publicationDate | 2024-06-05 | |
| heal.recordProvider | School of Science and Technology, MSC in Smart Cities and Communities | en_US |
| heal.type | masterThesis | en_US |
Αρχεία
Πρωτότυπος φάκελος/πακέτο
1 - 1 από 1
Φόρτωση...
- Ονομα:
- ThesisReport_MUSTAPHA_Akeem.pdf
- Μέγεθος:
- 2.45 MB
- Μορφότυπο:
- Adobe Portable Document Format
- Περιγραφή:
Φάκελος/Πακέτο αδειών
1 - 1 από 1
Δεν υπάρχει διαθέσιμη μικρογραφία
- Ονομα:
- license.txt
- Μέγεθος:
- 2.58 KB
- Μορφότυπο:
- Item-specific license agreed upon to submission
- Περιγραφή: