| Title |
Monthly Electric Power Sales Forecasting Algorithm Using LSTM-XGBoost Ensemble Model |
| DOI |
https://doi.org/10.5370/KIEE.2024.73.5.766 |
| Keywords |
Electric Power Sales; Power Consumption; Forecasting; LSTM; XGBoost; Ensemble Model |
| Abstract |
Electric power sales forecasting is important from various perspectives, including the nation, power producers, and sellers. At the national level, it serves as foundational information for electricity supply efficiency, energy policy establishment, and integration of renewable energy. Power producers utilize it for optimal operation and planning of power plants, and decisions on power plant expansion and investment. Sellers can use it for monthly energy balance evaluation, operational efficiency improvement, and financial analysis. In this paper, we analyze factors that affect electric power sales to predict medium and long-term electricity sales volume, and based on this, input variables that have a high correlation with monthly electricity sales volume are selected. Then, the LSTM-based deep neural network model and XGBoost model are learned using the selected data. We construct an ensemble model by combining the monthly forecasts of each model using a voting method and present an algorithm to predict electricity sales for the next 24 months. The proposed ensemble model showed improved performance over the prediction model using a single technique. |