Renewable Energy MicroGrid Power Forecasting: AI Techniques with Environmental Perspective

Amanul Islam^1*Fazidah Othman²

• ¹University of Colorado Colorado Springs, Colorado Springs, United States
• ²Universiti Malaya, Federal Territory of Kuala Lumpur, Malaysia

Abstract—Accurate power forecasting is a fundamental requirement for the reliable and sustainable operation of renewable energy–based microgrids, particularly under the inherent variability of solar and wind resources. This paper presents a comprehensive comparative study of traditional artificial intelligence models—including Artificial Neural Networks (ANN), Fuzzy Logic, and Adaptive Neuro-Fuzzy Inference Systems (ANFIS)—and advanced deep learning architectures such as Long Short-Term Memory (LSTM), Temporal Convolutional Networks (TCN), Transformer, and Squeeze-and-Excitation enhanced LSTM (SE+LSTM) for renewable power forecasting. Real-world hourly data collected from the King Saud University microgrid in Riyadh, Saudi Arabia, incorporating environmental variables such as solar irradiance, wind speed, temperature, humidity, and air pressure, are used for model training and evaluation. Forecasting performance is assessed using Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Sum of Squared Errors (SSE) across short-term (1-hour ahead) and mid-term (6-hour ahead) forecasting horizons. The results demonstrate that attention-based models significantly outperform conventional approaches, with the SE+LSTM model achieving the lowest RMSE of 0.7015 kW and MAPE of 2.01%, followed closely by the Transformer model. Statistical significance testing confirms that these improvements are not due to random variation. Overall, the findings highlight the critical role of environmental context in enhancing forecasting accuracy and demonstrate that attention-enhanced deep learning models provide a robust and environmentally informed framework for intelligent and sustainable microgrid energy management.