Short-Term Residential Load Forecasting Framework Based on Spatial-Temporal Fusion Adaptive Gated Graph Convolution...

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Accurate electric load prediction is vital for stable power grid operation.
• Conventional deep learning methods struggle with the spatio-temporal complexities of residential load data.
• Existing spatial graph representations are often limited, hindering inter-household learning.

Purpose of the Study:

• To propose a novel framework, Spatial-Temporal fusion adaptive gated graph convolution networks (STFAG-GCNs), for enhanced residential short-term load forecasting (STLF).
• To address the limitations of conventional methods in capturing both temporal dependencies and spatial structures in load data.
• To improve the dynamic modeling of spatio-temporal correlations for more accurate load predictions.

Main Methods:

• Development of a Spatial-Temporal fusion graph construction to capture unreflected correlations.
• Introduction of a gated adaptive fusion graph convolution (AFG-Conv) mechanism for dynamic spatio-temporal modeling.
• Integration of a gated temporal convolutional network (Gated TCN) with multiple STFGCNs in a unified layer to handle long sequences.

Main Results:

• STFAG-GCN demonstrated superior accuracy and robustness in real-world STLF datasets.
• The proposed framework significantly outperformed existing state-of-the-art methods.
• Ablation experiments confirmed the effectiveness and superiority of the STFAG-GCN components.

Conclusions:

• STFAG-GCN offers a significant advancement in short-term residential load forecasting.
• The novel fusion graph and adaptive convolution mechanisms effectively model complex spatio-temporal dynamics.
• The framework provides a more accurate and robust solution for power grid management.