Short-Term Price Forecasting Based on Feature Selection and Bidirectional LSTM

In the transformation of electricity market liberalization, the price sequence has gradually become a time series with high volatility and complexity. Relying solely on past experience to select feature sets is no longer suitable for the rapid fluctuations and complexities of electricity prices in different electricity markets, making it difficult to achieve ideal prediction accuracy. In order to enhance the accuracy of electricity price prediction, this paper proposes a method for electricity price prediction based on feature engineering and long short-term memory neural networks. Firstly, Copula entropy is used to simplify the calculation of information entropy. By maximizing the correlation and minimizing redundancy in the feature set, features with strong correlation to electricity prices are extracted. Then, a prediction model is constructed using Bi-LSTM to learn the long-term dependencies in the electricity price sequence information and obtain the predicted electricity price results. The proposed algorithm performs well in terms of prediction accuracy on the electricity price dataset in the PJM market. CBFS also outperforms other methods in terms of noise tolerance.