Enhancing Microgrid Performance Prediction with Attention-Based Deep Learning Models

Power oscillations in microgrid systems, driven by the variability of renewable energy sources like solar and wind, pose significant operational challenges. These oscillations cause grid instability, synchronization difficulties, and inefficient energy storage management, undermining microgrid reliability and efficiency. Addressing these issues requires precise load forecasting for predictive control, optimal resource allocation, stability maintenance, and cost-effective operation. This research proposes an integrated strategy utilizing convolutional and Gated Recurrent Unit (GRU) layers to extract temporal data from energy datasets, with an attention layer highlighting significant features for optimized load forecasting. Anchored by a Multi-Layer Perceptron (MLP) model, the framework handles comprehensive load forecasting and abnormal grid behavior identification. Evaluated with the Micro-grid Tariff Assessment Tool dataset, the model achieved an MAE of 0.39, RMSE of 0.28, and an R2-score of 98.89% in load forecasting, with nearperfect zero state prediction accuracy ( 99.9%). The proposed model outperforms conventional machine learning models such as support vector regression and random forest regression, making it particularly suitable for real-time applications and enhancing microgrid management. This abstract defines the problem, methodology, results, and significance of the research concisely and coherently.