Short-term Electricity Price Forecasting Method Based on CNN-LSTM by Random Forest and EEMD

Electricity spot prices are characterized by high volatility, making the selection of relevant variables and the construction of model crucial for accurate price forecasting. To better explore the correlation between industrial sector load profiles and spot prices, and to uncover hidden patterns within the data, we propose a short-term electricity price forecasting model designed to enhance forecasting accuracy. This model integrates Random Forest (RF), Ensemble Empirical Mode Decomposition (EEMD), Convolutional Neural Networks (CNN), and Long Short-Term Memory Networks (LSTM). Specifically, the RF algorithm is employed to identify industrial sector load data that are highly correlated with spot prices. EEMD is utilized to decompose the price data, thereby reducing volatility and instability. The chosen load data and the decomposed price sequences are then combined to form the input dataset for the CNN-LSTM network. Finally, the forecasting outputs from each component are reconstructed to produce the final prediction. Results demonstrate that the proposed model achieves superior prediction accuracy compared to other approaches.