A load forecasting method based on edge graph attention network

Area of Science:

• Artificial Intelligence
• Data Science
• Electrical Engineering

Background:

• Traditional load forecasting methods struggle with complex multi-dimensional feature relationships.
• Accurate power load forecasting is crucial due to increasing energy demands.

Purpose of the Study:

• To propose an innovative method for high-accuracy power load forecasting.
• To transform time series data into graph features for enhanced prediction capabilities.

Main Methods:

• Converting time series data into graph features by constructing a time-node graph structure.
• Utilizing the Edge Graph Attention Network (EGAT) to integrate node and edge feature information.
• Comparing EGAT against Gated Recurrent Units (GRU), Multi-Layer Perceptron networks (MLP), and Long Short-Term Memory (LSTM) models.

Main Results:

• The EGAT model demonstrated effectiveness in identifying significant features and understanding intricate temporal patterns.
• EGAT showed superior performance compared to GRU, MLP, and LSTM in load forecasting accuracy.
• The approach highlights strong potential for predicting energy demand with enhanced precision.

Conclusions:

• The proposed graph-based EGAT method significantly improves power load forecasting accuracy.
• Future research should address computational costs and optimize graph design for large-scale applications.
• The method offers a promising direction for advanced energy demand prediction systems.