A CEEMDAN–BiLSTM–GRU hybrid framework for assessing the impact of wind power generation on electricity prices: A comparative study of DK1 and DK2

Electricity price forecasting is critical for electricity markets with high renewable energy penetration, where increasing wind generation introduces significant volatility and uncertainty into price formation. Although deep learning and signal decomposition methods have shown promising results, studies jointly addressing forecasting accuracy, explainability, and robustness remain limited, particularly for the DK1 and DK2 regions of the Danish electricity market. This study proposes a hybrid CEEMDAN–BiLSTM–GRU framework for short-term electricity price forecasting. Electricity price series are first decomposed into intrinsic mode functions using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). Each component is then modeled using a hybrid Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (GRU) architecture, and the component forecasts are aggregated to obtain the final prediction. The proposed framework was evaluated using hourly electricity price and generation data covering the period 2014–2020. Results showed that the proposed model outperformed all benchmark models, achieving RMSE values of 2.9367 €/MWh and 4.1556 €/MWh, MAE values of 1.6010 €/MWh and 1.9288 €/MWh, and R² values of 0.9717 and 0.9559 for DK1 and DK2, respectively. Wilcoxon signed-rank tests confirmed that the performance improvements were statistically significant. SHAP-based analyses identified wind generation as the most influential variable affecting model predictions, while robustness analyses demonstrated stable performance during the COVID-19 period. The findings indicate that the proposed framework provides an accurate, robust, and interpretable solution for short-term electricity price forecasting in renewable-intensive electricity markets.