Enhancing Microgrid Performance Prediction with Attention-based Deep Learning Models

TL;DR

The paper tackles microgrid stability and the need for accurate load forecasting and anomaly detection under volatile generation. It introduces a hybrid CNN-GRU-Attention-MLP architecture that uses 1D convolution for temporal feature extraction, a GRU for sequential modeling, and an attention mechanism to weight informative time steps, with a final MLP head for regression/classification. The authors demonstrate strong performance on the Micro-grid Tariff Assessment Tool dataset, reporting $MAE=0.39$, $RMSE=0.28$, and $r^2=98.89%$ in load forecasting, along with approximately $99.9%$ zero-state accuracy, outpacing SVR, RF, and other baselines and aided by Shapley-value interpretations of feature importance. The results suggest the approach is suitable for real-time microgrid management, offering accurate forecasts, anomaly detection, and interpretable insights that can guide operational decisions.

Abstract

In this research, an effort is made to address microgrid systems' operational challenges, characterized by power oscillations that eventually contribute to grid instability. An integrated strategy is proposed, leveraging the strengths of convolutional and Gated Recurrent Unit (GRU) layers. This approach is aimed at effectively extracting temporal data from energy datasets to improve the precision of microgrid behavior forecasts. Additionally, an attention layer is employed to underscore significant features within the time-series data, optimizing the forecasting process. The framework is anchored by a Multi-Layer Perceptron (MLP) model, which is tasked with comprehensive load forecasting and the identification of abnormal grid behaviors. Our methodology underwent rigorous evaluation using the Micro-grid Tariff Assessment Tool dataset, with Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (r2-score) serving as the primary metrics. The approach demonstrated exemplary performance, evidenced by a MAE of 0.39, RMSE of 0.28, and an r2-score of 98.89\% in load forecasting, along with near-perfect zero state prediction accuracy (approximately 99.9\%). Significantly outperforming conventional machine learning models such as support vector regression and random forest regression, our model's streamlined architecture is particularly suitable for real-time applications, thereby facilitating more effective and reliable microgrid management.

Figures (8)

• Figure 1: The microgrid components and its structure.
• Figure 2: Architecture of the GRU.
• Figure 3: Attention unit mechanism.
• Figure 4: The proposed method architecture
• Figure 5: The performance of the model for predicted versus real value.