|
Forecasting total power generation in Thailand using time series and machine learning approaches |
|---|---|
| รหัสดีโอไอ | |
| Title | Forecasting total power generation in Thailand using time series and machine learning approaches |
| Creator | Taksaporn Triratpaladol |
| Contributor | Jirachai Buddhakulsomsiri, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Power generation forecasting, Time-series forecasting, Machine learning, Artificial neural network, Ensemble model |
| Abstract | This study develops and evaluates forecasting models for Thailand’s total monthly electricity generation, measured in gigawatt-hours (GWh), using data from January 1986 to December 2024. The research aims to examine the long-term trend and seasonal characteristics of national electricity generation, compare the forecasting accuracy of conventional statistical models and a machine learning approach, and assess whether an ensemble forecasting strategy improves predictive performance. Four forecasting methods are considered: Holt-Winters Exponential Smoothing, Seasonal Autoregressive Integrated Moving Average (SARIMA), Artificial Neural Network (ANN), and an Ensemble model based on averaging the three individual models' forecasts. Model performance is evaluated primarily using Mean Absolute Percentage Error (MAPE), with Mean Absolute Deviation (MAD) and Root Mean Square Error (RMSE) as supplementary measures of accuracy. The results show that Thailand’s monthly electricity generation exhibits a strong upward trend and clear multiplicative seasonality, reflecting rising electricity demand and recurring seasonal variation. In the holdout evaluation, the SARIMA and ANN models produced highly similar forecasting accuracy, indicating that a well-specified statistical model can remain competitive with a machine learning approach for monthly aggregate time-series data. The Ensemble model achieved the lowest MAPE and MAD, demonstrating that combining structurally different forecasting methods can reduce individual model errors and improve overall predictive reliability. These findings suggest that the Ensemble model is the most suitable forecasting framework for Thailand’s monthly electricity generation and highlight the importance of matching forecasting methods to data characteristics. |
