One day ahead hourly solar radiation forecasting technique | |
| Author | Khongpol Poka |
| Call Number | AIT Thesis no.SE-22-02 |
| Subject(s) | Photovoltaic power systems Solar radiation--Forecasting |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering Sustainable Energy Transition |
| Publisher | Asian Institute of Technology |
| Abstract | Solar photovoltaics (PV) has penetrated significantly into power grid because it is environmentally friendly and has an attractive payback period. Nevertheless, intermittency and variability are serious concerns especially to the system operator who has to operate the power grid. So, this study aimed to develop an accurate solar radiation forecasting model that can help manage these issues. A deep review of literatures indicated that single solar forecasting model may provide moderate results, but have difficulties during seasonal and extreme situations. The assessment of many individual models for forecasting hourly radiation one day ahead for selected locations in Thailand gave valuable feedbacks and showed the well performing models and indicated possibilities for their further refinement. This research, then applied three concepts focusing on the data and model structure for day ahead hourly solar radiation forecast. Fast Fourier transform (FFT) is a decomposing technique to transform data into frequency domain, so model could sense seasonal behavior from different frequency. To detect extreme events, K-means was implemented to split data into two group to distinguish normal and unexpected situations, such as sudden rain. Then, the combination of three good performing models were adopted because they are skillful in their own way, and so improved the results eventually. As data highly influences the performance of models, hourly data over a year composed of solar radiation, ambient temperature, relative humidity, and wind speed were preprocessed by three methods including IQR detection, limitation by clearness index, and Pearson correlation, to ensure data quality. The results show that the proposed model outperforms single models in terms of standard metrics - coefficient of determination (R2), mean bias error (MBE), and root mean square error (RMSE). It is approximately 30% better than individual models. Moreover, FFT and K-means are powerful because they lead the model to efficiently predict during the three different seasons, namely summer, rainy, and winter. Peak time is caught precisely by the model. The predicted hourly solar radiation can be converted to daily solar PV output through ‘Temperature correction empirical model’, The PV output by the proposed model compared with PVGIS, a commercial software, indicates the superior accuracy of the model. |
| Year | 2022 |
| Type | Thesis |
| School | School of Environment, Resources, and Development (SERD) |
| Department | Department of Energy and Climate Change (Former title: Department of Energy, Environment, and Climate Change (DEECC)) |
| Academic Program/FoS | Sustainable Energy Transition (SE) |
| Chairperson(s) | Kumar, Sivanappan |
| Examination Committee(s) | Singh, Jai Govind;Salam, P. Abdul |
| Scholarship Donor(s) | Her Majesty the Queen’s Scholarship (Thailand) |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2022 |