Short-Term Electric Load Forecasting in Thermal Power Plant Using AI and ML

Abstract

We examined the effectiveness of machine-learning-based electrical demand forecasting frameworks in supporting short-term operational planning for power generation facilities. To this end, a forecasting workflow was designed that integrates statistical learning methods with deep neural architectures to capture both temporal demand dynamics and exogenous weather influences. Model performance was assessed under controlled experimental conditions using multiple accuracy metrics, alongside sensitivity analyses to evaluate the influence of engineered features on predictive stability. The system employed a coordinated multi-model training approach, incorporating temporal decomposition, contextual feature construction, and climate-aware inputs to improve robustness across varying load profiles. Component-level ablation experiments were conducted to isolate the contribution of individual architectural and feature-engineering elements to overall forecasting accuracy. Results indicate that precise short-term load estimation extends beyond historical consumption modeling; it enables more efficient fuel scheduling, supports grid reliability, and enhances the system’s capacity to respond to real-time demand fluctuations.