Short-Term Aggregated Residential Load Forecasting of Low-Voltage Distribution Networks Based on Graph Neural...

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Advanced metering infrastructure generates vast amounts of fine-grained residential load data.
• Accurate aggregated load forecasting is crucial for power grid stability and electricity customer interactions.
• Existing methods often overlook spatial correlations in electricity consumption behavior.

Purpose of the Study:

• To propose a novel short-term aggregated residential load forecasting method.
• To address the limitations of existing methods by incorporating spatial correlations.
• To improve the accuracy of aggregated residential load predictions.

Main Methods:

• Utilized K-means clustering to group customers based on electricity consumption similarity.
• Constructed a spatial-temporal graph series incorporating customer groups and load profiles.
• Applied adaptive spatial-temporal synchronous graph convolutional networks for forecasting.

Main Results:

• The proposed method demonstrated significant improvements in forecasting accuracy.
• Experimental results on a real-world dataset validated the effectiveness of the approach.
• Outperformed several traditional benchmark forecasting methods.

Conclusions:

• The integration of spatial correlations via graph neural networks enhances load forecasting accuracy.
• The proposed K-means clustering and GNN-based method offers a more robust solution for aggregated residential load forecasting.
• This approach provides a valuable tool for power network operators to improve grid management.