LAUSR

Abstract In deregulated electricity markets, reliable electricity market price forecasting is the foundation for making the bidding strategy, operating dispatch control, and hedging volatility risk. However, electricity prices are high-volatile, nonstationary, multi-seasonal, making it difficult to estimate future trends. This paper proposes a hybrid model integrating a deep learning model, feature extraction and feature selection method to forecast short-term electricity prices. In the proposed framework, the ensemble empirical mode decomposition (EEMD) filter is utilized for multi-dimensional sequences, solving hidden characteristic extraction problems. The constructed feature space is identified and ranked under the max-dependency and min-redundancy (MRMR) criterion, improving the accuracy of feature selection. Finally, combining EEMD and MRMR with bidirectional long short-term memory (BiLSTM), a new hybrid framework is designed to improve the efficiency of short-term electricity price forecasting. Case studies on the PJM and New South Wales electricity markets confirm that our model outperforms alternatives on the forecasting accuracy. The average mean absolute percentage error (MAPE) of the proposed model is reduced by 4% to 21% compared to state-of-the-art models for 1-h and 24-h ahead forecasting. The proposed model has achieved relatively higher stability and adaptability in different forecasting steps and can better capture sophisticated fluctuations in electricity prices.