|
Medium-term electricity load demand forecasting from Provincial Electricity Authority’s monthly aggregated 96-interval data across four regions with LightGBM and XGBoost |
|---|---|
| รหัสดีโอไอ | |
| Title | Medium-term electricity load demand forecasting from Provincial Electricity Authority’s monthly aggregated 96-interval data across four regions with LightGBM and XGBoost |
| Creator | Chanatip Pongsangangan |
| Contributor | Rujira Chaysiri, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Electricity load forecasting, Medium-term forecasting, Machine learning, LightGBM, XGBoost, Gradient boosting, Thai electricity demand |
| Abstract | This research presents a comparative study of two machine learning models, Light Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting Machine (XGBoost), for medium-term electricity load forecasting medium-term electricity load demand using aggregated 96-interval monthly data from the Provincial Electricity Authority. The historical load data from 2017 to 2024 are divided into the Central, Northern, North-eastern and Southern regions of Thailand. The training data from 2017 to 2023 and the test data from 1st January to 31st December 2024 are applied. The forecasted data is classified into day types including Peak-day, Workday, Saturday, Sunday and Holiday. The forecasting models are built using lag-based features such as the same interval from the previous month and the interval from the year before, along with rolling averages. Both models forecast the 96-interval monthly load curve using lag-based feature representations. Model performance is evaluated using mean absolute percentage error. The results show that XGBoost delivers stronger performance in MAPE than LightGBM in all regions except the Southern region. The forecasting framework developed in this study can support PEA with more reliable monthly planning and operational decision-making. |
