Enhanced Transformer-BiLSTM Deep Learning Framework for Day-Ahead Energy Price Forecasting

Accurate forecasting of energy prices is critical to effectively mitigate operational risks and make strategic bidding decisions in day-ahead (DA) electricity markets. However, it is highly challenging due to volatile characteristics, seasonality, rapid spikes, and other nonlinear factors of price signals. In the given context, deep learning (DL) has gained attention in recent years due to its high potential in nonlinear approximation, but each model has its strengths and limitations. Therefore, this paper proposes a hybrid DL approach for time-series DA energy price forecasting based on the Transformer and Bidirectional Long Short-Term Memory (BiLSTM) model that facilitates strategically combining various components to extract patterns and further improve the sequence processing task. The proposed model uses a transformer architecture to capture patterns, temporal dynamics, and BiLSTM networks to forecast energy price fluctuations. The proposed approach is validated with simulations based on price data from the New York Independent System Operator, and the results show that the proposed approach consistently outperforms state-of-the-art DL models, achieving the lowest MAE (2.7818${\$}$/MWh), RMSE (6.4937${\$}$/MWh), sMAPE (6.6060%), MAPE (6.3741%) and highest R$^{2}$(0.9393). The effectiveness of the proposed approach was justified through various case studies from different perspectives.