Volume 64

Adaptive and Interpretable Electricity Price Forecasting via Sliding-Update BiLSTM with Attention Qichen Wang, Zhengmeng Hou, Yilin Guo, Liangchao Huang, Tianle Shi1, Tian Zhang, Nan Cai

Abstract

Electricity price forecasting is highly challenging due to market volatility, abrupt spikes, and recurrent seasonal patterns. This study proposes an adaptive and interpretable deep learning framework, termed Sliding-Update Bidirectional LSTM with Attention (SU-BiLSTM-AM), which combines a dynamic update mechanism, bidirectional sequence modeling, and attention-based feature weighting to improve accuracy and robustness under volatile market conditions. Using day-ahead prices from the European Energy Exchange (EEX), SU-BiLSTM-AM consistently outperforms recurrent and sequence-learning baselines, reducing forecast errors by 14–23% compared with LSTM. The model demonstrates robustness under extreme regimes and maintains reasonable computational cost, making it suitable for intelligent energy systems, bidding strategies, and risk management in volatile electricity markets.

Keywords Electricity price forecasting, bidirectional LSTM, attention mechanism, intelligent energy systems, renewable integration, time series