| Author: | Liao, Xuan |
| Title: | Development of machine learning-based approach for solar potential estimation |
| Advisors: | Wong, Man Sing (LSGI) |
| Degree: | Ph.D. |
| Year: | 2024 |
| Subject: | Solar energy Photovoltaic power systems Machine learning Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Land Surveying and Geo-Informatics |
| Pages: | xiv, 131 pages : color illustrations |
| Language: | English |
| Abstract: | The global pursuit of carbon neutrality aims to mitigate greenhouse gas emissions and establish a sustainable future. Solar energy is one of the most promising approaches, as it produces minimal greenhouse gas emissions. Installing solar photovoltaic panels on rooftops maximizes solar irradiation reception while reducing energy transmission losses and costs. Given the high installation costs of solar photovoltaic (PV) panels, accurately estimating solar potential to determine optimal installation locations is crucial to ensure economic benefits exceed installation costs, making the investment viable. Accurate solar potential estimation faces several challenges: i) Various natural (such as clouds and weather) and artificial factors (building, morphological features) affect solar irradiation, making quantification and establishing non-linear relationships challenging. ii) Current algorithms struggle with the spatio-temporal characteristics of solar irradiation, which is crucial for effective estimation. iii) Large-scale solar potential estimation involves processing vast data, posing computational challenges. This study employs a hierarchical assessment framework based on machine learning to estimate solar potential, including physical and geographical potential. The major achievements of this thesis are: (1) Four machine learning models (Gradient Boosting Machine (GBM), Random Forest (RF), Support Vector Regression, Multilayer Perceptron) were used to estimate land surface solar irradiation in Australia, China, and Japan using meteorological data, Himawari-8 satellite cloud and aerosol products, and solar observation data. GBM showed the highest accuracy, suggesting its effectiveness for large regions and applicability globally with similar datasets. This method generated accurate and continuous solar maps to display solar resource distributions at large-scale regions. (2) To address geographic heterogeneity in estimating land surface solar irradiation, the Dual-gate Temporal Fusion Transformer (DGTFT) was proposed. Applied to datasets from Australia, China, and Japan, the proposed network outperformed traditional machine learning methods, with a minimum Coefficient of determination (R2) increase of 23.88%, Mean Absolute Error (MAE) decrease of 43.18%, and Normalized Root Mean Square Error (nRMSE) decrease of 62.79%. These results suggest that the proposed network not only improves estimation performance but also provides interpretable results for understanding the network mechanism. (3) This study proposes a parametric-based data and model dual-driven method to estimate annual rooftop solar irradiation at a fine spatial resolution. Three machine learning methods (RF, GBM, and AdaBoost) were cross-compared based on R2, MAE, and computation time. In a Hong Kong case study, RF outperformed GBM and AdaBoost, with R2=0.77 and MAE=22.83 kWh/m2/year. Training and prediction time for rooftop solar irradiation was within 13 hours, achieving a 99.32% reduction compared to the physical-based hemispherical viewshed algorithm, indicating the proposed method's accuracy and speed for large datasets. (4) The DGTFT model was employed to estimate hourly rooftop solar irradiation, capturing spatio-temporal distribution variations. The proposed method achieved highly accurate results, with R2=0.90, MAE=26.90 MJ/m2, RMSE=32.39 MJ/m2, and was 56 times faster than the model-driven method. These results demonstrate the high spatio-temporal resolution rooftop solar maps' reliability for solar potential assessment. This thesis offers promising approaches for estimating solar potential from physical to geographical potential at high spatio-temporal resolution, utilizing Geographic Information System (GIS) representation of multi-source data and exploring non-linear relationships using Geospatial Artificial Intelligence (GeoAI) methods. The findings provide a reliable reference for planning and installing solar PV systems. |
| Rights: | All rights reserved |
| Access: | open access |
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/200/13389