A Novel Approach to Electricity Price Prediction: LDS-GRU Algorithm for Long-Tail Effects

As China's electricity spot trading market evolves, accurate electricity price forecasting has become crucial for optimizing power system operations and reducing costs and risks for market participants. Traditional forecasting methods struggle with the nonlinear and volatile nature of electricity prices, particularly with extreme values or long-tail effects. This paper proposes an advanced forecasting method that integrates the Label Distribution Smoothing (LDS) algorithm with the Gated Recurrent Unit (GRU) model, termed LDS-GRU, to enhance prediction accuracy by addressing long-tail effects. We conduct an empirical analysis of intraday electricity prices in the Guangdong power market for 2023, revealing significant time-domain characteristics and variability in price distributions. By segmenting the data into 24 hourly models and applying LDS for preprocessing, our approach demonstrates substantial improvements in forecasting performance compared to traditional GRU models. The findings underscore the importance of accounting for extreme price values and provide a more accurate solution for electricity price forecasting, with clear guidance for market traders.