Prediction of electricity demand in weakly interconnected power systems using an ensemble time series model with a Bayesian Optimization approach Academic Article in Scopus uri icon

abstract

  • Accurate short-term electricity demand forecasting is essential for effective operational planning, especially in power systems with limited interconnection, constrained transmission infrastructure, and increasing shares of renewable generation. This study proposes an integrated framework that combines Bayesian Optimization (BO) with two-time series models, FB Prophet, and SARIMA, to enhance nodal-level forecast accuracy. BO is employed both for model-specific hyperparameter tuning and to determine optimal ensemble weights, enabling a data-driven and adaptive approach to demand prediction. The resulting forecasts are incorporated into a multi-objective dispatch optimization model that minimizes total operating costs and CO2 emissions, creating a seamless pipeline from prediction to operational decision-making. The methodology is applied to Region 8 of the Mexican National Electric System (SEN), a region characterized by weak interconnection and operational constraints. Results indicate that the BO-optimized ensemble reduces the Mean Absolute Error (MAE) by up to 52% compared to individual models. When integrated into the dispatch framework, these forecasts support strategies that balance economic and environmental objectives. The proposed framework is generalizable and scalable, offering a valuable tool for data-driven planning in other weakly interconnected or renewable-intensive systems, with clear implications for supporting energy transition strategies under uncertainty. © 2025 Institution of Chemical Engineers

publication date

  • August 1, 2025