ABSTRACT Electricity price forecasting is crucial for power market operations and improved trading decisions. Short‐term electricity price primarily depends on demand and type of generation. High integration of intermittent renewable energy generation intensifies price volatility. Deep learning models are increasingly employed to model the nonlinear and volatile dynamics of electricity prices, addressing the limitations of traditional statistical forecasting methods. However, they struggle to extract electricity price features with apparent regularity. To overcome these limitations, this paper develops a novel hybrid short‐term electricity price forecasting model that integrates multi‐size depthwise convolutional neural networks (MDCNN), bidirectional gated recurrent units (BiGRU), and multi‐head attention mechanism. The MDCNN component employs varied convolutional kernel sizes to extract both short and long‐term information from input variables, alleviating information overload. These features are then processed by BiGRU, which captures intricate non‐linear temporal dependencies. This proposed hybrid model is augmented by a multi‐head attention mechanism to dynamically prioritize crucial features, improving interpretability and robustness. Finally, this model is fine‐tuned by Bayesian optimization to obtain optimal hyperparameters. Advanced data‐preprocessing techniques, such as two‐stage dimensionality reduction approach and outlier treatment by random forest‐based imputation, are applied to ensure data quality. For a case study of Spanish electricity market, the proposed model outperforms benchmark models with an of 3.31/MWh, of 2.34/MWh, and score of 91.2%. Additionally, Friedman rank test and post hoc analysis confirm the model's statistical significance. These results demonstrate the model's effectiveness in capturing complex price patterns and its potential practical applications in the power market.
Prajesh et al. (Thu,) studied this question.