Outlier-adaptive-based non-crossing quantiles method for day-ahead electricity price forecasting

In deregulated electricity markets, accurate and reliable day-ahead electricity price forecasting (EPF) is beneficial for hedging volatility risks, implementing dispatch controls, and formulating bidding strategies. As the electricity price series is characterized by high volatility, non-stationarity and multi-seasonality, predicting its future trend poses a challenge. Moreover, most existing forecasting models lack tolerance for outliers disturbance, which is insufficient to cover the demands of actual wholesale market. To this end, an outlier-adaptive-based non-crossing quantiles regression (OANQR) model is proposed for day-ahead EPF. In this model, to tackle disturbances from price outliers, local outlier factor (LOF) module is implemented for detecting and isolating abnormal price. Then, the filtered non-stationary price series is decomposed into sub-sequence using variational mode decomposition (VMD) module, reducing complexity and randomness. In terms of the temporal variability, a non-crossing quantiles regression is proposed by integrating cross-conformal symmetry and skip connection recurrent neural network to enhance the reliability and sharpness of the forecasted electricity price intervals. Finally, through validation on real data from different electricity markets, the study reveals that the developed OANQR surpasses existing models with respect to deterministic and interval forecasting of day-ahead EPF.