Genetic Algorithm-Based Model for Short-Term Load Forecasting in Isolated Microgrids

Isolated microgrids face operational challenges due to restricted generation capacity and high sensitivity to consumption fluctuations. Reliable short-term forecasting is essential to support decision-making in these constrained environments. Accurate short-term load forecasting plays a key role in the planning and operation of electrical power systems, especially in isolated microgrids with limited generation capacity. This study proposes a hybrid forecasting model that combines the Recursive Least Squares algorithm with the K-Nearest Neighbors method, enhanced through the application of Genetic Algorithm optimization techniques. The model integrates weather conditions, calendar variables, and historical consumption data to identify behavioral patterns and improve forecast performance. The implementation was carried out in MATLAB version 2024rd, using the Global Optimization Toolbox structures for Genetic Algorithm and Direct Search methods to fine-tune the model parameters. Real operational data from 2023 and 2024, collected from isolated electrical systems serving tourist areas, were used to validate the proposed model. The results show that the hybrid approach outperforms classical Recursive Least Squares and Artificial Neural Network models, particularly during periods of high demand variability. This improved forecasting capacity supports energy providers in making informed decisions for scheduling maintenance and operational actions, while ensuring efficient power generation at reduced costs. The methodology is adaptable to other isolated or small-scale power systems and contributes to improving the reliability and cost-effectiveness of energy planning in similar regional contexts.