Real-Time Electricity Price Forecasting with Dynamic Graph Convolution and Adaptive Temporal Fusion
Abstract
INTRODUCTION: Real-time electricity price forecasting is critical for real-time dispatch, automatic generation control, spot trading and risk management in electricity markets.
Objectives
Since large-scale wind and photovoltaic renewable energy grid integration brings strong random fluctuation of renewable power output, real-time electricity prices exhibit stronger volatility, higher randomness and tighter time coupling than day-ahead prices; the randomness of renewable power is one core factor causing abrupt price jumps, leading conventional forecasting models struggle to capture dynamic spatial–temporal dependencies and sudden fluctuation patterns.
Methods
To address these issues, this paper proposes a real-time electricity price forecasting model based on dynamic graph convolution (DGC), adaptive gated temporal fusion, and Transformer–BiGRU hybrid structure. The model uses a dynamic graph convolution module to model real-time changing topological correlations among power system features, introduces an adaptive gated fusion mechanism to enhance the weighting of real-time key features, and integrates Transformer and bidirectional GRU to capture both long-range global dependencies and high-frequency local fluctuations.
Results
Experiments are conducted on real-time electricity price and real-time operation data from Shandong electricity market. Results show that the proposed model achieves average prediction accuracy of 88.2%, with MAE and MSE reduced to 0.071 and 0.009 respectively, which outperforms all baseline methods significantly.
Conclusion
The proposed method provides high-precision support for real-time market transactions, real-time scheduling and intelligent decision-making.