Abstract
Aim/Background: The increasing integration of renewable energy sources into microgrids underscores the need for accurate demand forecasting to ensure stable and efficient microgrid operation. However, the inherent unpredictability of renewable energy sources presents significant challenges in energy management. This study aims to develop and validate an AI-driven demand forecasting model that improves prediction accuracy compared to traditional methods, thereby enhancing energy management in renewable microgrids.
Methods: A hybrid forecasting model that combines long short-term memory (LSTM) networks and Convolutional Neural Networks (CNN) was proposed. The model leverages historical energy consumption and meteorological data for training, ensuring robust and accurate predictions. Data preprocessing, training, and validation were performed meticulously to evaluate performance. The proposed model was compared with traditional forecasting techniques, including ARIMA and Exponential Smoothing, to assess its accuracy.
Results: The hybrid LSTM-CNN model demonstrated superior performance, achieving an R2 value of 0.87, a Mean Absolute Error (MAE) of 1.45 MWh, and a Root Mean Squared Error (RMSE) of 2.12 MWh. These results significantly outperform conventional forecasting methods, such as ARIMA and exponential smoothing, highlighting the model’s enhanced accuracy and ability to address key challenges in renewable energy forecasting.
Conclusion: This study establishes the effectiveness of the hybrid LSTM-CNN approach in terms of improving the demand forecasting of renewable microgrids. The model’s superior accuracy provides a reliable tool for real-time decision-making, energy distribution optimization, and cost reduction. Policymakers and energy stakeholders can use these insights to develop sustainable energy systems, with future research focusing on scaling up the model and exploring behavior and market pricing to improve forecasting precision.

Key words: AI-driven demand forecasting, Long Short-Term Memory (LSTM) networks; Renewable microgrids; Energy management; Convolutional Neural Networks (CNNs); Forecasting accuracy