MONTHLY ELECTRICITY CONSUMPTION FORECASTING METHOD BASED ON ARIMA-RF AND CEEMDAN-SSA DECOMPOSITION TECHNIQUE

• المؤلف الرئيسي: Husni Mubarak, Muhammad
• التنسيق: Theses
• اللغة: Indonesia
• الوصول للمادة أونلاين: https://digilib.itb.ac.id/gdl/view/61875

Electricity is one of the most vital forms of energy in everyday life. This fact triggers an increase in the demand for electrical power from year to year. To ensure the supply of electric power remains safe, reliable, and cost-effective, electrical energy consumption forecasting is becoming an important process in electric power system planning. Monthly electricity consumption could be affected by a range of factors and often contains linear and non-linear patterns. Therefore, traditional forecasting methods are deemed insufficient as it does not consider these factors, and this method is not suitable for forecasting non-linear time series. This study uses the hybrid Autoregressive Integrated Moving Average (ARIMA) – Random Forest (RF) method combined with the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) - Singular Spectrum Analysis (SSA) decomposition technique to overcome the problems that have been mentioned. The CEEMDAN technique decomposes electricity consumption into components with different characteristics, while SSA is used to help reduces noise. The ARIMA model is then used to predict components with linear characteristics, and Random Forest predicts components with non-linear characteristics. The experiments were carried out on four categories of consumers in Bali Province, namely households, businesses, industries, and other categories. Economic, weather, and tourism factors are used as input variables to forecast electricity consumption in the four categories of consumers. The test results show that compared to SARIMA, supports vector regression (SVR), and extreme gradient boosting (XGB), the proposed method can significantly improve forecasting accuracy with an average MAPE of 3.33%.