A data driven model based approach for medium-to-long-term electricity price forecasting in power markets

Area of Science:

• Energy Economics
• Computational Intelligence
• Time Series Analysis

Background:

• Accurate medium-to-long-term electricity price forecasting is crucial for market participants' bidding strategies and cost mitigation.
• Existing forecasting models face challenges with high-dimensional data, poor adaptability, and price volatility.

Purpose of the Study:

• To propose a data-driven approach for accurate medium-to-long-term electricity price forecasting.
• To enhance forecasting model adaptability and reduce data dimensionality and price volatility.

Main Methods:

• Key datasets (historical electricity, coal, natural gas prices) were selected using decision tree importance assessment.
• Data sequences were denoised and volatility reduced using Fast Fourier Transformation (FFT).
• A Gray Wolf Optimization (GWO)-CNN-LSTM-Attention model was developed for forecasting.

Main Results:

• The proposed FFT-GWO-CNN-LSTM-Attention (FGCLA) algorithm achieved significant forecasting accuracy improvements.
• FGCLA demonstrated average accuracy gains of 57.21% over LSTM and 49.69% over Transformer.
• The algorithm effectively reduced forecasting errors in medium-to-long-term electricity price predictions.

Conclusions:

• The FGCLA algorithm offers improved dimension reduction, adaptability, and volatility suppression for electricity price forecasting.
• The developed model provides a robust solution for accurate medium-to-long-term electricity price prediction.
• This approach can aid market participants in optimizing strategies and managing expenditures effectively.