<![CDATA[Accurate and robust short-term electricity load forecasting is essential for reliable power system operation, particularly in tropical regions where demand is strongly influenced by nonlinear consumption patterns and weather-induced uncertainty. Conventional statistical models often struggle to capture these characteristics, while standalone neural networks may suffer from training instability and sensitivity to initialization. This study proposes a hybrid soft computing framework that integrates fuzzy logic–based weather uncertainty representation, genetic algorithm–driven optimization, and artificial neural networks (Fuzzy–GA–ANN) for short-term load forecasting. The fuzzy component provides an uncertainty-aware abstraction of meteorological effects, while the genetic algorithm enhances training robustness by mitigating local minima and initialization sensitivity. The framework is evaluated using a large-scale hourly load dataset from the Java–Bali interconnected power system, covering multiple operational horizons (1-hour, 6-hour, and day-ahead). Experimental results demonstrate that the proposed model consistently outperforms classical statistical baselines (ETS and SARIMA) and ANN-based variants across all horizons. The most significant improvements are observed for day-ahead forecasting, where the proposed approach achieves substantially lower forecasting errors and improved training stability. These findings indicate that combining uncertainty-aware feature representation with robust optimization yields reliable and operationally viable forecasting performance in climate-sensitive power systems.]]>
Copyrights © 2025
