Accurate short-term forecasting of renewable energy sources is essential for stable and efficient microgrid operation. Existing models primarily focus on either solar or wind prediction, often neglecting their combined stochastic behavior within isolated systems. This study presents a comparative evaluation of three machine-learning models—Random Forest, ANN, and LSTM—for short-term solar and wind forecasting in microgrid environments. Historical meteorological data and power generation records are used to train and validate three ML models: Random Forest, Long Short-Term Memory, and Artificial Neural Networks. Each model is optimized to capture nonlinear and rapidly fluctuating weather dynamics. Forecasting performance is quantitatively evaluated using Mean Absolute Error, Root Mean Square Error, and Mean Percentage Error. The predicted values are integrated into a microgrid energy management system to enhance operational decisions such as battery storage scheduling, diesel generator coordination, and load balancing. Among the evaluated models, the ANN achieved the lowest prediction error with an MAE of 64.72 kW on the one-year dataset, outperforming both LSTM and Random Forest. The novelty of this study lies in integrating multi-source data into a unified ML-based predictive framework, enabling improved reliability, reduced fossil fuel usage, and enhanced energy resilience in remote microgrids. This research used Orange 3.40 software and Python 3.12 code for prediction. By enhancing forecasting accuracy, the project seeks to reduce reliance on fossil fuels, lower operational costs, and improve grid stability. Outcomes will provide scalable insights for remote microgrids transitioning to renewables.
Patel et al. (Thu,) studied this question.